Artificial Intelligence for Speech Recognition

ARTIFICIAL INTELLIGENCE FOR SPEECH RECOGNITIONIRFANA TABASSUM T ABSTRACT Artificial intelligence (AI) for speech recognition involves two basic ideas. First, it involves studying the thought processes of human beings. Second, it deals with representing those processes via machines (like computers, robots, etc).AI is behavior of a machine, which, if performed by a human being, would be called intelligence. It makes machines smarter and more useful, and is less expensive than natural intelligence. Natural language processing (NLP) refers to artificial intelligence methods of communicating with a computer in a natural language like English. The main objective of a NLP program is to understand input and initiate action. The input words are scanned and matched against internally stored known words. Identification of a keyword causes some action to be taken. In this way, one can communicate with the computer in one's language CONTENTS 1. Introduction 1.1 Problems 1.2 Tools 1.3 Applications 1 2 7 11 12 15 16 17 19 19 21 25 26 28 2. The Technology 3. Speech recognition 3.1 Speech recognition process 3.2 Terms and concepts 4. Speaker independency 4.1 Speaker Dependence Vs Speaker Independence 4.2 System Configuration 5. Working of the system 5.1 Speaker- dependent word recognizer 5.2 What software is available? 5.3 What technical issues need to be considered when purchasing this system? 28 5.4 How does the technology differ from other technologies?29 5.5 What factors need to be considered when using speech recognition technology? 30 6. The limits of speech recognition 7. Application 7.1 Health care 7.2 Military 7.3 Training air traffic controllers 7.4 Telephony and other domains 31 34 34 35 38 39 39 39 41 42 7.5 People with disabilities 7.6 Hands-free computing 8. Conclusion 9. Bibliography Artificial intelligence for speech recognition Chapter 1 INTRODUCTION The speech recognition process is performed by a software component known as the speech recognition engine. The primary function of the speech recognition engine is to process spoken input and translate it into text that an application understands. The application can then do one of two things: The application can interpret the result of the recognition as a command. In this case , the application is a command and control application. If an application handles the recognized text simply as text, then it is considered a dictation application. The user speaks to the computer through a microphone, which in turn, identifies the meaning of the words and sends it to NLP device for further processing. Once recognized, the words can be used in a variety of applications like display, robotics, commands to computers, and dictation. No special commands or computer language are required. There is no need to enter programs in a special language for creating software. Voice XML takes speech recognition even further. Instead of talking to your computer, you're essentially talking to a web site, and you're doing this over the phone. OK , you say, well, what exactly is speech recognition? Simply put, it is the process of converting spoken input to text. Speech recognition is thus sometimes referred to as speech-to-text .Speech recognition allows you to provide input to an application with your voice. Just like clicking with your mouse, typing on your keyboard, or pressing a key on the phone keypad provides input to an application; speech recognition allows you to provide input by talking. In the desktop world, you need a microphone to be able to do this. In the Voice XML world, all you need is a telephone. When you dial the telephone number of a big company, you are likely to hear the sonorous voice of a cultured lady who responds to your call with great courtesy saying ³welcome to company X. Please give me the extension number you want´ .You pronounce the extension number, your name, and the name of the person you want to contact. If the called person accepts the call, the connection is given quickly. This is artificial intelligence where an automatic callhandling system is used without employing any telephone operator. AI is the study of the abilities for computers to perform tasks, which currently are better done by humans. AI has an interdisciplinary field where computer science intersects with philosophy, psychology, engineering and other fields. Humans make decisions based upon 1 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition experience and intention. The essence of AI in the integration of computer to mimic this learning process is known as Artificial Intelligence Integration. 1.1 Problems The general problem of simulating (or creating) intelligence has been broken down into a number of specific sub-problems. These consist of particular traits or capabilities that researchers would like an intelligent system to display. The traits described below have received the most attention. 1.1.1 Deduction, reasoning, problem solving Early AI researchers developed algorithms that imitated the step-by-step reasoning that humans were often assumed to use when they solve puzzles, play board games or make logical deductions. By the late 1980s and '90s, AI research had also developed highly successful methods for dealing with uncertain or incomplete information, employing concepts from probability and economics. For difficult problems, most of these algorithms can require enormous computational resources ² most experience a "combinatorial explosion": the amount of memory or computer time required becomes astronomical when the problem goes beyond a certain size. The search for more efficient problem solving algorithms is a high priority for AI research. Human beings solve most of their problems using fast, intuitive judgments rather than the conscious, step-by-step deduction that early AI research was able to model. AI has made some progress at imitating this kind of "sub-symbolic" problem solving: embodied agent approaches emphasize the importance of sensorimotor skills to higher reasoning; neural net research attempts to simulate the structures inside human and animal brains that gives rise to this skill. 1.1.2 Knowledge representation Knowledge representation and knowledge engineering are central to AI research. Many of the problems machines are expected to solve will require extensive knowledge about the world. Among the things that AI needs to represent are: objects, properties, categories and 2 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition relations between objects situations, events, states and time causes and effects knowledge about knowledge (what we know about what other people know) and many other, less well researched domains. A complete representation of "what exists" is an ontology (borrowing a word from traditional philosophy), of which the most general are called upper ontologies. 1.1.2.1 Among the most difficult problems in knowledge representation are: y Default reasoning and the qualification problem Many of the things people know take the form of "working assumptions." For example, if a bird comes up in conversation, people typically picture an animal that is fist sized, sings, and flies. None of these things are true about all birds. John McCarthy identified this problem in 1969 as the qualification problem: for any commonsense rule that AI researchers care to represent, there tend to be a huge number of exceptions. Almost nothing is simply true or false in the way that abstract logic requires. AI research has explored a number of solutions to this problem. y The breadth of commonsense knowledge The number of atomic facts that the average person knows is astronomical. Research projects that attempt to build a complete knowledge base of commonsense knowledge (e.g., Cyc) require enormous amounts of laborious ontological engineering ² they must be built, by hand, one complicated concept at a time. A major goal is to have the computer understand enough concepts to be able to learn by reading from sources like the internet, and thus be able to add to its own ontology. y The sub symbolic form of some commonsense knowledge Much of what people know is not represented as "facts" or "statements" that they could actually say out loud. For example, a chess master will avoid a particular chess position because it "feels too exposed" or an art critic can take one look at a statue and instantly realize that it is a fake. These are intuitions or tendencies that are represented in the brain non-consciously and sub-symbolically. Knowledge like this informs, supports and provides a context for symbolic, conscious knowledge. As with the related problem of sub-symbolic reasoning, it is hoped that situated AI or computational intelligence will provide ways to represent this kind of knowledge. 3 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition 1.1.3 Planning Intelligent agents must be able to set goals and achieve them. They need a way to visualize the future (they must have a representation of the state of the world and be able to make predictions about how their actions will change it) and be able to make choices that maximize the utility (or "value") of the available choices. In classical planning problems, the agent can assume that it is the only thing acting on the world and it can be certain what the consequences of its actions may be. However, if this is not true, it must periodically check if the world matches its predictions and it must change its plan as this becomes necessary, requiring the agent to reason under uncertainty. Multi-agent planning uses the cooperation and competition of many agents to achieve a given goal. Emergent behavior such as this is used by evolutionary algorithms and swarm intelligence. 1.1.4 Learning Machine learning has been central to AI research from the beginning. Unsupervised learning is the ability to find patterns in a stream of input. Supervised learning includes both classification and numerical regression. Classification is used to determine what category something belongs in, after seeing a number of examples of things from several categories. Regression takes a set of numerical input/output examples and attempts to discover a continuous function that would generate the outputs from the inputs. In reinforcement learning the agent is rewarded for good responses and punished for bad ones. These can be analyzed in terms of decision theory, using concepts like utility. The mathematical analysis of machine learning algorithms and their performance is a branch of theoretical computer science known as computational learning theory 1.1.5 Natural language processing Figure : 1.1 ASIMO uses sensors and intelligent algorithms to avoid obstacles and navigate stairs. 4 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition Natural language processing gives machines the ability to read and understand the languages that humans speak. Many researchers hope that a sufficiently powerful natural language processing system would be able to acquire knowledge on its own, by reading the existing text available over the internet. Some straightforward applications of natural language processing include information retrieval (or text mining) and machine translation. 1.1.6 Motion and manipulation Figure : 1.2 The Care-Providing robot FRIEND uses sensors like cameras and intelligent algorithms to control the manipulator in order to support disabled and elderly people in their daily life activities.The field of robotics is closely related to AI. Intelligence is required for robots to be able to handle such tasks as object manipulation and navigation, with sub-problems of localization (knowing where you are), mapping (learning what is around you) and motion planning (figuring out how to get there) 1.1.7 Perception Machine perception is the ability to use input from sensors (such as cameras, microphones, sonar and others more exotic) to deduce aspects of the world. Computer vision is the ability to analyze visual input. A few selected subproblems are speech recognition, facial recognition and object recognition. 5 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition 1.1.8 Social intelligence Figure : 1.3 Kismet, a robot with rudimentary social skills Emotion and social skills play two roles for an intelligent agent. First, it must be able to predict the actions of others, by understanding their motives and emotional states. (This involves elements of game theory, decision theory, as well as the ability to model human emotions and the perceptual skills to detect emotions.) Also, for good human-computer interaction, an intelligent machine also needs to display emotions. At the very least it must appear polite and sensitive to the humans it interacts with. At best, it should have normal emotions itself. 1.1.9 Creativity Figure : 1.4 TOPIO, a robot that can play table tennis, developed by TOSY. A sub-field of AI addresses creativity both theoretically (from a philosophical and psychological perspective) and practically (via specific implementations of systems that generate outputs that can be considered creative). A related area of computational research is Artificial Intuition and Artificial Imagination. 6 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition 1.2 Tools In the course of 50 years of research, AI has developed a large number of tools to solve the most difficult problems in computer science. A few of the most general of these methods are discussed below. 1.2.1 Search and optimization Many problems in AI can be solved in theory by intelligently searching through many possible solutions Reasoning can be reduced to performing a search. For example, logical proof can be viewed as searching for a path that leads from premises to conclusions, where each step is the application of an inference rule. Planning algorithms search through trees of goals and subgoals, attempting to find a path to a target goal, a process called means-ends analysis. Robotics algorithms for moving limbs and grasping objects use local searches in configuration space. Many learning algorithms use search algorithms based on optimization. Simple exhaustive searches are rarely sufficient for most real world problems: the search space (the number of places to search) quickly grows to astronomical numbers. The result is a search that is too slow or never completes. The solution, for many problems, is to use "heuristics" or "rules of thumb" that eliminate choices that are unlikely to lead to the goal (called "pruning the search tree"). Heuristics supply the program with a "best guess" for what path the solution lies on. A very different kind of search came to prominence in the 1990s, based on the mathematical theory of optimization. For many problems, it is possible to begin the search with some form of a guess and then refine the guess incrementally until no more refinements can be made. These algorithms can be visualized as blind hill climbing: we begin the search at a random point on the landscape, and then, by jumps or steps, we keep moving our guess uphill, until we reach the top. Other optimization algorithms are simulated annealing, beam search and random optimization. Evolutionary computation uses a form of optimization search. For example, they may begin with a population of organisms (the guesses) and then allow them to mutate and recombine, selecting only the fittest to survive each generation (refining the guesses). Forms of evolutionary computation include swarm intelligence algorithms (such as ant colony or particle 7 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition swarm optimization) and evolutionary algorithms (such as genetic algorithms and genetic programming). 1.2.2 Logic Logic was introduced into AI research by John McCarthy in his 1958 Advice Taker proposal. Logic is used for knowledge representation and problem solving, but it can be applied to other problems as well. For example, the satplan algorithm uses logic for planning and inductive logic programming is a method for learning. Several different forms of logic are used in AI research. Propositional or sentential logic is the logic of statements which can be true or false. First-order logic also allows the use of quantifiers and predicates, and can express facts about objects, their properties, and their relations with each other. Fuzzy logic, is a version of first-order logic which allows the truth of a statement to be represented as a value between 0 and 1, rather than simply True (1) or False (0). Fuzzy systems can be used for uncertain reasoning and have been widely used in modern industrial and consumer product control systems. Subjective logic models uncertainty in a different and more explicit manner than fuzzy-logic: a given binomial opinion satisfies belief + disbelief + uncertainty = 1 within a Beta distribution. By this method, ignorance can be distinguished from probabilistic statements that an agent makes with high confidence. Default logics, non-monotonic logics and circumscription are forms of logic designed to help with default reasoning and the qualification problem. Several extensions of logic have been designed to handle specific domains of knowledge, such as: description logics situation calculus, event calculus and fluent calculus (for representing events and time) causal calculus belief calculus; and modal logics. 1.2.3 Probabilistic methods for uncertain reasoning Many problems in AI (in reasoning, planning, learning, perception and robotics) require the agent to operate with incomplete or uncertain information. Starting in the late 80s and early 90s, Judea Pearl and others championed the use of methods drawn from probability theory and economics to devise a number of powerful tools to solve these problems. Bayesian networks are a very general tool that can be used for a large number of problems: reasoning (using the Bayesian inference algorithm), learning (using the expectation8 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition maximization algorithm), planning (using decision networks) and perception (using dynamic Bayesian networks). Probabilistic algorithms can also be used for filtering, prediction, smoothing and finding explanations for streams of data, helping perception systems to analyze processes that occur over time (e.g., hidden Markov models or Kalman filters). A key concept from the science of economics is "utility": a measure of how valuable something is to an intelligent agent. Precise mathematical tools have been developed that analyze how an agent can make choices and plan, using decision theory, decision analysis, information value theory. These tools include models such as Markov decision processes, dynamic 1.2.4 Classifiers and statistical learning methods The simplest AI applications can be divided into two types: classifiers ("if shiny then diamond") and controllers ("if shiny then pick up"). Controllers do however also classify conditions before inferring actions, and therefore classification forms a central part of many AI systems. Classifiers are functions that use pattern matching to determine a closest match. They can be tuned according to examples, making them very attractive for use in AI. These examples are known as observations or patterns. In supervised learning, each pattern belongs to a certain predefined class. A class can be seen as a decision that has to be made. All the observations combined with their class labels are known as a data set. When a new observation is received, that observation is classified based on previous experience. A classifier can be trained in various ways; there are many statistical and machine learning approaches. The most widely used classifiers are the neural network, kernel methods such as the support vector machine, k-nearest neighbor algorithm, Gaussian mixture model, naive Bayes classifier, and decision tree. The performance of these classifiers have been compared over a wide range of tasks. Classifier performance depends greatly on the characteristics of the data to be classified. There is no single classifier that works best on all given problems; this is also referred to as the "no free lunch" theorem. Determining a suitable classifier for a given problem is still more an art than science. 9 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition 1.2.5 Neural networks Figure : 1.5 A neural network is an interconnected group of nodes, akin to the vast network of neurons in the human brain.The study of artificial neural networks began in the decade before the field AI research was founded, in the work of Walter Pitts and Warren McCullough. Other important early researchers were Frank Rosenblatt, who invented the perceptron and Paul Werbos who developed the back propagation algorithm. The main categories of networks are acyclic or feed forward neural networks (where the signal passes in only one direction) and recurrent neural networks (which allow feedback). Among the most popular feed forward networks are perceptrons, multi-layer perceptrons and radial basis networks. Among recurrent networks, the most famous is the Hopfield net, a form of attractor network, which was first described by John Hopfield in 1982. Neural networks can be applied to the problem of intelligent control (for robotics) or learning, using such techniques as Hebbian learning and competitive learning. Jeff Hawkins argues that research in neural networks has stalled because it has failed to model the essential properties of the neocortex, and has suggested a model (Hierarchical Temporal Memory) that is based on neurological research. 1.2.6 Control theory Control theory, the grandchild of cybernetics, has many important applications, especially in robotics. 1.2.7 Languages AI researchers have developed several specialized languages for AI research, including Lisp and Prolog. 10 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition 1.3 Applications Artificial intelligence has successfully been used in a wide range of fields including medical diagnosis, stock trading, robot control, law, scientific discovery, video games, toys, and Web search engines. Frequently, when a technique reaches mainstream use, it is no longer considered artificial intelligence, sometimes described as the AI effect .It may also become integrated into artificial life. 1.3.1Competitions and prizes There are a number of competitions and prizes to promote research in artificial intelligence. The main areas promoted are: general machine intelligence, conversational behavior, data-mining, driverless cars, robot soccer and games. 1.3.2 Platforms A platform (or "computing platform")is defined by Wikipedia as "some sort of hardware architecture or software framework (including application frameworks), that allows software to run." As Rodney Brooks pointed out many years ago, it is not just the artificial intelligence software that defines the AI features of the platform, but rather the actual platform itself that affects the AI that results, i.e, we need to be working out AI problems on real world platforms rather than in isolation. A wide variety of platforms has allowed different aspects of AI to develop, ranging from expert systems, albeit PC-based but still an entire real-world system to various robot platforms such as the widely available Roomba with open interface. 11 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition Chapter 2 THE TECHNOLOGY A human identity recognition system based on voice analysis could have seamless applications. The ASR (Automatic Speaker Recognition) is one such system. Automatic Speaker Recognition is a system that can recognize a person based on his/her voice. This is achieved by implementing complex signal processing algorithms that run on a digital computer or a processor.This Application is analogous to the fingerprint recognition system or other biometrics recognition systems that are based on certain characteristics of a person. There are several occasions when we want to identify a person from a given group of people even when the person is not present for physical examination. For example, when a person converses on a telephone, all we have is the person¶s voice for analysis. It then makes sense to develop a recognition system based on voice. Speaker recognition has typically been classified as either a verification or identification task. Speaker verification is usually the simpler of the two since it involves the comparison of the input signal with a single given stored reference pattern. Therefore, the verification task only requires a system to verify, if the speaker is the same as the person he/she identifies himself/herself. Speaker identification is more complex because the test speaker must be compared against a number of reference speakers to determine if a match can be made. Not only the input signal is to be examined to see if it came from a speaker, but the identification of the individual speaker is also necessary. The identification of speakers remains a difficult task for a number of reasons. First, the acquiring of a unique speech signal can suffer as a result of the variation of the voice inputs from a speaker and environmental factors. Both the volume and pace of speech can vary from one test to another. Also, unless initially constrained, an extensive vocabulary or unstructured grammar can affect results. Background noise must also be kept to the minimum so that a changing environment will not divert the speaker¶s attention or the final voicing of a word or sentence. As a result, many restrictions and clarifications have been placed on speaker and speech recognition systems. 12 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition One such restriction involves using a closed set for speaker recognition. A closed set implies that only speakers within the original stored set will be asked to be identified. An open set would allow the extra possibility of a test speaker not coming from the initially trained set of speakers, thereby requiring the system to recognize the speaker as not belonging to the original set. An open set system may also have the task to learning a new speaker and placing him or her within the original set for future reference. Another common restriction involves using a test dependent speaker recognition system. This type of system would require the speaker to utter a unique word or phrase to be compared against the original set of like phrases. Text-independent recognition, which for most cases is more complex and difficult to perform, identifies the speaker regardless of the text or phrase spoken. Once an utterance, or signal, has been recorded, it is usually necessary to process it to get the voiced signal in a form that makes classification and recognition possible. Various methods have included the use of power spectrum values, spectrum coefficients, linear predictive coding, and a nearest neighbour distance algorithm. Tests have also shown that although spectrum coefficients and linear predictive coding have given better results for conventional template and statistical classification methods, power spectrum values have performed better when using neural networks during the final recognition stages. Various methods have also been used to perform the classification and recognition of the processed speech signal. Statistical methods utilizing Hidden Markov Models, linear vector quantifiers, or classical techniques such as template matching have produced encouraging, yet limited success. Recent deployments using neural networks, while producing varied success rates, have offered more options regarding the types of inputs sent to the networks, as well as provided the ability to learn speakers in both an off and online manner. Although backpropagation networks have traditionally been used, the implementation of more sophisticated networks, such as an ART 2 network, has been made. ASR can be broadly classified into four types: 1. 2. Text-independent identification Text-independent verification 13 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition 3. 4. Text-dependent identification Text-dependent verification Speaker identification is a procedure by which a speaker is identified from a group of µn¶ people. It should be noted that a totally new speaker not belonging to the group could wrongly be identified as someone from within the group.Speaker verification is a procedure by which a speaker who claims his/her identity is verified as being correct or not. A fundamental requirement for any ASR system is gathering reference samples and finding certain features from the voice that are characteristic to a person. These feature vectors are then stored. When a new test sample is made available, the references are either searched to find the closest match (in case of identification), or a threshold of a distance measure is checked (in case of verification). The next aspect to the considered is text-dependency. In a text-independent situation, the reference utterance and the test utterance are not the same. This type of recognition system finds its applications in criminology. In a text-dependent situation, the reference utterance and the test utterance are the same, which gives us a higher degree of accuracy. This type of recognition system has applications where security is a matter of concern, such as access to a building to a lab, to a computer, etc. The dominant technology used in ASR is called the Hidden Markov Model, or HMM. This technology recognizes speech by estimating the likelihood of each phoneme at contiguous, small regions (frames) of the speech signal. Each word in a vocabulary list is specified in terms of its component phonemes. A search procedure is used to determine the sequence of phonemes with the highest likelihood. This search is constrained to only look for phoneme sequences that correspond to words in the vocabulary list, and the phoneme sequence with the highest total likelihood is identified with the word that was spoken. In standard HMMs, the likelihoods are computed using a Gaussian Mixture Model; in the HMM/ANN framework, these values are computed using an artificial neural network (ANN). 14 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition Chapter 3 SPEECH RECOGNITION The user speaks to the computer through a microphone, which in turn, identifies the meaning of the words and sends it to NLP device for further processing. Once recognized, the words can be used in a variety of applications like display, robotics, commands to computers, and dictation. The word recognizer is a speech recognition system that identifies individual words. Early pioneering systems could recognize only individual alphabets and numbers. Today, majority of word recognition systems are word recognizers and have more than 95% recognition accuracy. Such systems are capable of recognizing a small vocabulary of single words or simple phrases. One must speak the input information in clearly definable single words, with a pause between words, in order to enter data in a computer. Continuous speech recognizers are far more difficult to build than word recognizers. You speak complete sentences to the computer. The input will be recognized and, then processed by NLP. Such recognizers employ sophisticated, complex techniques to deal with continuous speech, because when one speaks continuously, most of the words slur together and it is difficult for the system to know where one word ends and the other begins. Unlike word recognizers, the information spoken is not recognized instantly by this system. What is a speech recognition system? A speech recognition system is a type of software that allows the user to have their spoken words converted into written text in a computer application such as a word processor or spreadsheet. The computer can also be controlled by the use of spoken commands. Speech recognition software can be installed on a personal computer of appropriate specification. The user speaks into a microphone (a headphone microphone is usually supplied with the product). The software generally requires an initial training and enrolment process in order to teach the software to recognize the voice of the user. A voice profile is then produced that is unique to that individual. This procedure also helps the user to learn how to µspeak¶ to a computer. 15 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition 3.1 Speech recognition process DISPLAY APPLICATION S SPEECH RECOGNITIO N DEVICE DICTATING USER COMMANDS TO COMPUTER INPUT TO OTHER ROBOTS, EXPERTS NLP VIOCE SOUND UNDERSTAND ING DIALOGUE WITH USER Figure:3.1 After the training process, the user¶s spoken words will produce text; the accuracy of this will improve with further dictation and conscientious use of the correction procedure. With a well-trained system, around 95% of the words spoken could be correctly interpreted. The system can be trained to identify certain words and phrases and examine the user¶s standard documents in order to develop an accurate voice file for the individual. However, there are many other factors that need to be considered in order to achieve a high recognition rate. There is no doubt that the software works and can liberate many learners, but the process can be far more time consuming than first time users may appreciate and the results can often be poor. This can be very demotivating, and many users give up at this stage. Quality support from someone who is able to show the user the most effective ways of using the software is essential. 16 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition When using speech recognition software, the user¶s expectations and the advertising on the box may well be far higher than what will realistically be achieved. µYou talk and it types¶ can be achieved by some people only after a great deal of perseverance and hard work. 3.2 Terms and Concepts Following are a few of the basic terms and concepts that are fundamental to speech recognition. It is important to have a good understanding of these concepts when developing Voice XML applications. 3.2.1 Utterances When the user says something, this is known as an utterance. An utterance is any stream of speech between two periods of silence. Utterances are sent to the speech engine to be processed. Silence, in speech recognition, is almost as important as what is spoken, because silence delineates the start and end of an utterance. Here's how it works. The speech recognition engine is "listening" for speech input. When the engine detects audio input - in other words, a lack of silence -- the beginning of an utterance is signaled. Similarly, when the engine detects a certain amount of silence following the audio, the end of the utterance occurs. Utterances are sent to the speech engine to be processed. If the user doesn¶t say anything, the engine returns what is known as a silence timeout - an indication that there was no speech detected within the expected timeframe, and the application takes an appropriate action, such as reprompting the user for input. An utterance can be a single word, or it can contain multiple words (a phrase or a sentence). 3.2.2 Pronunciations The speech recognition engine uses all sorts of data, statistical models, and algorithms to convert spoken input into text. One piece of information that the speech recognition engine uses to process a word is its pronunciation, which represents what the speech engine thinks a word should sound like. Words can have multiple pronunciations associated with them. For example, the word ³the´ has at least two pronunciations in the U.S. English language: ³thee´ and ³thuh.´ 17 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition As a Voice XML application developer, you may want to provide multiple pronunciations for certain words and phrases to allow for variations in the ways your callers may speak them. 3.2.3 Grammars As a Voice XML application developer, you must specify the words and phrases that users can say to your application. These words and phrases are defined to the speech recognition engine and are used in the recognition process. You can specify the valid words and phrases in a number of different ways, but in Voice XML, you do this by specifying a grammar. A grammar uses a particular syntax, or set of rules, to define the words and phrases that can be recognized by the engine. A grammar can be as simple as a list of words, or it can be flexible enough to allow such variability in what can be said that it approaches natural language capability. 3.2.4 Accuracy The performance of a speech recognition system is measurable. Perhaps the most widely used measurement is accuracy. It is typically a quantitative measurement and can be calculated in several ways. Arguably the most important measurement of accuracy is whether the desired end result occurred. This measurement is useful in validating application design Another measurement of recognition accuracy is whether the engine recognized the utterance exactly as spoken. Another measurement of recognition accuracy is whether the engine recognized the utterance exactly as spoken. This measure of recognition accuracy is expressed as a percentage and represents the number of utterances recognized correctly out of the total number of utterances spoken. It is a useful measurement when validating grammar design. Recognition accuracy is an important measure for all speech recognition applications. It is tied to grammar design and to the acoustic environment of the user. You need to measure the recognition accuracy for your application, and may want to adjust your application and its grammars based on the results obtained when you test your application with typical users. 18 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition Chapter 4 SPEAKER INDEPENDENCY The speech quality varies from person to person. It is therefore difficult to build an electronic system that recognizes everyone¶s voice. By limiting the system to the voice of a single person, the system becomes not only simpler but also more reliable. The computer must be trained to the voice of that particular individual. Such a system is called Speaker-dependent system. Speaker-independent system can be used by anybody, and can recognize any voice, even though the characteristics vary widely from one speaker to another. Most of these systems are costly and complex. Also, these have very limited vocabularies. It is important to consider the environment in which the speech recognition system has to work. The grammar used by the speaker and accepted by the system, noise level, noise type, position of the microphone, and speed and manner of the user¶s speech are some factors that may affect the quality of the speech recognition. 4.1 Speaker Dependence Vs Speaker Independence Speaker Dependence describes the degree to which a speech recognition system requires knowledge of a speaker¶s individual voice characteristics to successfully process speech. The speech recognition engine can ³learn´ how you speak words and phrases; it can be trained to your voice. Speech recognition systems that require a user to train the system to his/her voice are known as speaker-dependent systems. If you are familiar with desktop dictation systems, most are speaker dependent. Because they operate on very large vocabularies, dictation systems perform much better when the speaker has spent the time to train the system to his/her voice. Speech recognition systems that do not require a user to train the system are known as speaker-independent systems. Speech recognition in the VoiceXML world must be speakerindependent. Think of how many users (hundreds, maybe thousands) say be calling into your web site. You cannot require that each caller train the system to his or her voice. The speech recognition system in a voice-enabled web application MUST successfully process the speech of 19 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition many different callers without having to understand the individual voice characteristics of each caller. 4.1.1 Advantages of speaker independent system The advantage of a speaker independent system is obvious anyone can use the system without first training it. However, its drawbacks are not so obvious. One limitation is the work that goes into creating the vocabulary templates. To create reliable speaker independents templates, someone must collect and process numerous speech sample. This is a time-consuming task; creating these templates is not a one-time effort. Speaker-independent templates are language-dependant, and the templates are sensitive not only to two dissimilar languages but also to the differences between British and American English. Therefore, as part of your design activity, you would need to create a set of templates for each language or a major dialect that your customers use. Speaker independent systems also have a relatively fixed vocabulary because of the difficulty in creating a new template in the field at the user¶s site. 4.1.2 The advantage of a speaker-dependent system A speaker dependent system requires the user a train the ASR system by providing examples of his own speech. Training can be tedious process, but the system has the advantage of using templates that refer only to the specific user and not some vague average voice. The result is language independence. You can say ja, si, or ya during training, as long as you are consistent. The drawback is that the speaker-dependent system must do more than simply match incoming speech to the templates. It must also include resources to create those templates. 4.1.3 Which is better: For a given amount of processing power, a speaker dependent system tends to provide more accurate recognition than a speaker independent system. A speaker independent system is not necessarily better: the difference in performance stems from the speaker independent template encompassing wide speech variations. 20 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition 4.2 System Configuration Figures 4.2.1 and 4.2.2 show the identification system and the verification system configuration, respectively .The first part of the system consists of the data acquisition hardware that acquires the speech, performs some signal conditioning, digitizes it and gives it to the computer/processor. The second part consists of core signal processing and system identification techniques to extract speaker specific features. These features are stored and are used at a later time for the actual identification/verification test. At this stage, the system is ready for identification or verification. Now, when the test sample is uttered by one of the members of the group, the speech is digitized and the features are extracted. For identification, distances between this vector and all the reference vectors are measured and the closest vector is picked up as the correct one. This vector would correspond to a person whom the system claims as having been identified. For verification, the person claims his/her identity. The distance between the corresponding reference vector and the test vector is the computed. If the measured distance is less than a set threshold, the verification system accepts the speaker; if not, it rejects the speaker. VOICE SAMPLE ADC WITH SIGNAL CONDITIONING ALGORITHM TO SELECT FEATURES MEASUREMENT OF DISTANCE DECISION MAKING REFERENCE VECTORS IDENTITY OF PERSON Figure 4.1: Speaker Identification 21 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition VOICE SAMPLE (PERSON CLAIMS HIS/HER IDENTITY ADC WITH SIGNAL CONDITIONIN G ALGORITHM TO SELECT FEATURES MEASUREMENT OF DISTANCE REFERENCE VECTOR OF THE SPEAKER PERSON VERIFIED OR NOT THRESHOLD COMPARISON Figure 4.2: Speaker Verification The voice input to the microphone produces an analogue speech signal. An analogue-todigital converter (ADC) converts this speech signal into binary words that are compatible with digital computer. The converted binary version is then stored in the system and compared with previously stored binary representations of words and phrases. The current input speech is compared one at a time with the previously stored speech pattern after searching by the computer. When a match occurs, recognition is achieved. The spoken word is binary form is written on a video screen or passed along to a natural language understanding processor for additional analysis. Since most recognition systems are speaker-dependent, it is necessary to train a system to recognize the dialect of each new user. During training, the computer displays a word and the user reads it aloud. The computer digitizes the user¶s voice and stores it. The speaker has to read aloud about 1,000 words. Based on these samples, the computer can predict how the user utters some words that are likely to be pronounced differently by different people. The block diagram of a speaker-dependent word recognizer is shown in Fig. 4.2.1 The user speaks before the microphone, which converts the sound into electrical signal. The electrical analogue signal from microphone is fed to an amplifier provided with automatic gain control (AGC) to produce an amplified output signal in a specific optimum voltage range, even when the input signal varies from feeble to loud. 22 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition The analogue signal, representing a spoken word, contains many individual frequencies of various amplitudes and different phases, which when blended together take the shape of a complex waveform . A set of filters is used to break this complex input signal into its component parts. Band pass filters (BEP) pass on frequencies only in certain frequency range, rejecting all other frequencies. Generally, about sixteen filters are used; a simple system may contain a minimum of three filters. The more the number of filters user, the higher the probability of accurate recognition. Presently, switched capacitor digital filters are used because these can be custom-built in integrated circuit form. These are smaller and cheaper than active filters using operational amplifiers. The filter output is then fed to the ADC to translate the analogue signal into digital word. The ADC samples the filter outputs many times a second. Each sample represents a different amplitude of the signal . Evenly spaced vertical lines represent the amplitude of the audio filter output at the instant of sampling. Each value is then converted to a binary number proportional to the amplitude of the sample. A central processor unit controls the input circuits that are fed by the ADCs. A large RAM stores all the digital values in a buffer area. This digital information, representing the spoken word, is now accessed by the CPU to process it further. The normal speech has a frequency range of 200 Hz to 7 kHz. Recognizing a telephone call is more difficult as it has bandwidth limitation of 300Hz to 3.3 Hz. As explained earlier, the spoken words are processed by the filters and ADCs. The binary representation of each of these words becomes a template or standard, against which the future words are compared. These templates are stored in the memory. Once the storing process is completed, the system can go into its active mode and is capable of identifying spoken words. As each word is spoken, it is converted into binary equivalent and stored in RAM. The computer then starts searching and compares the binary input pattern with the templates. It is to be noted that even if the same speaker talks the same text, there are always slight variations in amplitude or loudness of the signal, pitch, frequency difference, time gap, etc. Due to this reason, there is never a perfect match between the template and binary input word. The 23 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition pattern matching process therefore uses statistical techniques and is designed to look for the best fit. The values of binary input words are subtracted from the corresponding values, in the templates. If both the values are same, the difference is zero and there is perfect match. If not, the subtraction produces some difference or error. The smaller the error, the better the match. When the best match occurs the word is identified and displayed on the screen or used in some other manner. The search process takes a considerable amount of time as the CPU has to make many comparisons before recognition occurs. This necessitates use of very high-speed processors. A large RAM is also required as even though a spoken word may last only a few hundred milliseconds, but the same is translated into many thousands of digital words. It is important to not e that alignment of words and templates are to be matched correctly in time, before computing the similarity score. This process, termed as dynamic time warping, recognizes that different speaker pronounce the same words at different speeds as well as elongate different parts of the same word. This is important for the speaker-independent recognizers. Continuous speech recognizers are far more difficult to build than word recognizers. You can speak complete sentences to the computer. The input will be recognized and, when processed by NLP, understood. Such recognizers employ sophisticated, complex techniques to deal with continuous speech, because when one speaks continuously, most of the words slur together and it is difficult for the system to know where one word ends and the other begins. Unlike word recognizers, the information spoken is not recognized instantly by this system. 24 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition Chapter 5 WORKING OF THE SYSTEM The voice input to the microphone produces an analogue speech signal. An analogue to digital converter (ADC) converts this speech signal into binary words that are compatible with digital computer. The converted binary version is then stored in the system and compared with previously stored binary representation of words and phrases. The current input speech is compared one at a time with the previously stored speech pattern after searching by the computer. When a match occurs, recognition is achieved. The spoken word in binary form is written on a video screen or passed along to a natural language understanding processor for additional analysis. Since most recognition systems are speaker-dependent, it is necessary to train a system to recognize the dialect of each new user. During training, the computer displays a word and user reads it aloud. The computer digitizes the user¶s voice and stores it. The speaker has to read aloud about 1000 words. Based on these samples, the computer can predict how the user utters some words that are likely to be pronounced differently by different users. The block diagram of a speaker- dependent word recognizer is shown in figure. The user speaks before the microphone, which converts the sound into electrical signal. The electrical analogue signal from the microphone, is fed to an amplifier provided with automatic gain control (AGC) to produce an amplified output signal in a specific optimum voltage range, even when the input signal varies from feeble to loud. The analogue signal, representing a spoken word, contains many individual frequencies of various amplitudes and different phases, which when blended together take the shape of a complex wave form as shown in figure. A set of filters is used to break this complex signal into its component parts. Band pass filters (BFP) pass on frequencies only on certain frequency range, rejecting all other frequencies. Generally, about 16 filters are used; a simple system may contain a minimum of three filters. The more number of filters used, the higher the probability of accurate recognition. Presently, switched capacitor digital filters are used because these can be custom- built in integrated circuit form. These are smaller and cheaper than active filters using operational amplifiers. The filter output is then fed to the ADC to translate the analog signal into digital word. The ADC samples the filter output many times a second. Each sample represents different amplitude of the signal .A CPU controls the input circuits that are fed by the ADC¶s. A 25 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition large RAM stores all the digital values in a buffer area. This digital information, representing the spoken word, is now accessed by the CPU to process it further. 5.1 Speaker- dependent word recognizer I N P U T C I R C U I T S RAM DIGITISED SPEECH BPF ADC Figure 5.1 The normal speech has a frequency range of 200 Hz to 7KHz. Recognizing a telephone call is more difficult as it has bandwidth limitations of 300Hz to 3.3KHz.As explained earlier the spoken words are processed by the filters and ADCs. The binary representation of each of these word becomes a template or standard against which the future words are compared. These templates are stored in the memory. Once the storing process is completed, the system can go into its active mode and is capable of identifying the spoken words. As each word is spoken, it is converted into binary equivalent and stored in RAM. The computer then starts searching and compares the binary input pattern with the templates. It is to be noted that even if the same speaker talks the same text, there are always slight variations in amplitude or loudness of the signal, pitch, frequency difference, time gap etc. Due to this reason there is never a perfect match between the template and the binary input word. The pattern matching process therefore uses statistical techniques and is designed to look for the best fit. The values of binary input words are subtracted from the corresponding values in the templates. If both the values are same, the difference is zero and there is perfect match. If not, the 26 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition subtraction produces some difference or error. The smaller the error, the better the match. When the best match occurs, the word templates are to be matched correctly in time, before computing the similarity score. This process, termed as dynamic time warping recognizes that different speakers pronounce the same word at different is identified and displayed on the screen or used in some other manner. The search process takes a considerable amount of time, as the CPU has to make many comparisons before recognition occurs. This necessitates use of very high-speed processors. A Large RAM is also required as even though a spoken word may last only a few hundred milliseconds, but the same is translated into many thousands of digital words. It is important to note that alignment of words and speeds as well as elongate different parts of the same word. This is important for the speaker- independent recognizers. Now that we've discussed some of the basic terms and concepts involved in speech recognition, let's put them together and take a look at how the speech recognition process works. As you can probably imagine, the speech recognition engine has a rather complex task to handle, that of taking raw audio input and translating it to recognized text that an application understands. The major components discussed are: ‡ Audio input ‡ Grammar(s) ‡ Acoustic Model ‡ Recognized text The first thing we want to take a look at is the audio input coming into the recognition engine. It is important to understand that this audio stream is rarely pristine. It contains not only the speech data (what was said) but also background noise. This noise can interfere with the recognition process, and the speech engine must handle (and possibly even adapt to) the environment within which the audio is spoken. As we've discussed, it is the job of the speech recognition engine to convert spoken input into text. To do this, it employs all sorts of data, statistics, and software algorithms. Its first job is to process the incoming audio signal and convert it into a format best suited for further analysis. Once the speech data is in the proper format, the engine searches for the best match. It does this by taking into consideration the words and phrases it knows about (the active 27 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition grammars), along with its knowledge of the environment in which it is operating for Voice XML, this is the telephony environment). The knowledge of the environment is provided in the form of an acoustic model. Once it identifies the the most likely match or what was said, it returns what it recognized as a text string. Most speech engines try very hard to find a match, and are usually very "forgiving." But it is important to note that the engine is always returning it's best guess for what was said. 5.1.1 Acceptance and Rejection When the recognition engine processes an utterance, it returns a result. The result can be either of two states: acceptance or rejection. An accepted utterance is one in which the engine returns recognized text. Whatever the caller says, the speech recognition engine tries very hard to match the utterance to a word or phrase in the active grammar. Sometimes the match may be poor because the caller said something that the application was not expecting, or the caller spoke indistinctly. In these cases, the speech engine returns the closest match, which might be incorrect. Some engines also return a confidence score along with the text to indicate the likelihood that the returned text is correct. Not all utterances that are processed by the speech engine are accepted. Acceptance or rejection is flagged by the engine with each processed utterance. 5.2 What software is available? There are a number of publishers of speech recognition software. New and improved versions are regularly produced, and older versions are often sold at greatly reduced prices. Invariably, the newest versions require the most modern computers of well above average specification. Using the software on a computer with a lower specification means that it will run very slowly and may well be impossible to use. There are two main types of speech recognition software: discrete speech and continuous speech. Discrete speech software is an older technology that requires the user to speak one word at a time . Dragon Dictate Classic Version 3 is one example of discrete speech software, as it has fewer features, is simple to train and use and will work on Continuous speech software allows the user to dictate normally. In fact, it works best when it hears complete sentences, as it 28 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition interprets with more accuracy when it recognizes the context. The delivery can be varied by using short phrases and single words, following the natural pattern of speech. 5.3 What technical issues need to be considered when purchasing this system? The latest versions of speech recognition software (September 2001) require a Pentium 3 processor and 256 MB of memory. Currently, Dragon Naturally Speaking Version 4 and IBM Via Voice Millennium edition have been used in school settings. Very good results can be obtained with these on fast, high-memory machines. When purchasing a machine, it is worth mentioning to the supplier that it will be required for running speech recognition software. Whether choosing a desktop or portable computer, it will also require a good quality duplex (input and output) sound card. Poor sound quality will reduce the recognition accuracy. The microphones supplied with the software may be perfectly adequate, but better results can often be obtained by using a noise-cancelling microphone. In addition, mobile voice recorders allow a number of users to produce dictation that can be downloaded to the main speech recognition system, but be aware of some of the complexities of their use. 5.4 How does the technology differ from other technologies? Speech recognition systems produce written text from the user¶s dictation, without using, or with only minimal use of, a traditional keyboard and mouse. This is an obvious benefit to many people who, for any number of reasons, do not find it easy to use a keyboard, or whose spelling and literacy skills would benefit from seeing accurate text. The limitations to this type of software are that: It needs to be completely tailored to the user and trained by the user. It is often set up on one machine, and so can create difficulties for a user who works from many locations, for example from school and home. It depends on the user having the desire to produce text and be able to invest the time, training and perseverance necessary to achieve it. It is most successful for those competent in the art of dictation. 29 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition A speech recognition system is a powerful application in that the software¶s recognition of the user¶s voice pattern and vocabulary improves with use. A useful tip is to ensure that voice files can be backed up regularly. 5.5 What factors need to be considered when using speech recognition technology? The Becta SEN Speech Recognition Project describes the key factors to success as µThe Three Ts¶: Time, Technology and Training: Time Take time to choose the most appropriate software and hardware and match it to the user. One option for new users is to start with discrete speech software. The skills learned whilst using it can be transferred to more sophisticated speech recognition software. If the new user is unable to make effective use of discrete speech recognition software, then it is unlikely they will succeed with continuous speech software. Familiarisation with the product and frequent breaks between talking are also helpful.older computers. Training With speech recognition systems, both the software and the user require training. Patience and practice are required. The user needs to take things slowly, practicing putting their thoughts into words before attempting to use the system. Technology The best results are generally achieved using a high-specification machine. Sound cards and microphones are a key feature for success, as is access to technical support and advice. 30 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition Chapter 6 THE LIMITS OF SPEECH RECOGNITION To improve speech recognition applications, designers must understand acoustic memory and prosody. Continued research and development should be able to improve certain speech input, output, and dialogue applications. Speech recognition and generation is sometimes helpful for environments that are hands-busy, eyes-busy, mobility required, or hostile and shows promise for telephone-based ser-vices. Dictation input is increasingly accurate, but adoption outside the disabled-user community has been slow compared to visual interfaces. Obvious physical problems include fatigue from speaking continuously and the disruption in an office filled with people speaking. By understanding the cognitive processes surrounding human ³acoustic memory´ and processing, interface designers may be able to integrate speech more effectively and guide users more successfully. By appreciating the differences between human-human interaction and human-computer interaction, designers may then be able to choose appropriate applications for human use of speech with computers. The key distinction may be the rich emotional content conveyed by prosody, or the pacing, intonation, and amplitude in spoken language. The emotive aspects of prosody are potent for human interaction but may be disruptive for human-computer interaction. The syntactic aspects of prosody, such as rising tone for questions, are important for a system¶s recognition and generation of sentences. Now consider human acoustic memory and processing. Short-term and working Memory are some-times called acoustic or verbal mems the human brain that transiently holds chunks of information and solves problems also supports speaking and listening. Therefore, working on tough problems is best done in quiet environments without speaking or listening to someone. However, because physical activity is handled in another part of the brain, problem solving is compatible with routine physical activities like walking and driving. In short, humans speak and walk easily but find it more difficult to speak and think at the same time .Similarly when operating a computer, most humans type (or move a mouse) and think but find it more difficult to speak and think at the same time. Hand-eye coordination is accomplished in different brain structures, so typing or mouse movement can be performed in parallel with problem solving. 31 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition Product evaluators of an IBM dictation software the human brain that transiently holds chunks of information and solves problems also supports speaking and listening. Therefore, working on tough problems is best done in quiet environments²without speaking or listening to someone. However, because physical activity is handled in another part of the brain, problem solving is compatible with routine physical activities like walking and driving. In short, humans speak and walk easily but find it more difficult to speak and think at the same time . Similarly when operating a computer, most humans type (or move a mouse) and think but find it more difficult to speak and think at the same time. Hand-eye coordination is accomplished in different brain structures, so typing or mouse movement can be performed in parallel with problem solving. Product evaluators of an IBM dictation software package also noticed this phenomenon. They wrote that ³thought for many people is very closely linked to language. In keyboarding, users can continue to hone their words while their fingers output an earlier version. In dictation, users may experience more interference between outputting their initial thought and elaborating on it.´ Developers of commercial speech recognition software packages recognize this problem and often advise dictation of full paragraphs or documents, followed by a review or proofreading phase to correct errors. Since speaking consumes precious cognitive resources, it is difficult to solve problems at the same time. Proficient keyboard users can have higher levels of parallelism in problem solving while performing data entry. This may explain why after 30 years of ambitious attempts to provide military pilots with speech recognition in cockpits, aircraft designers persist in using hand-input devices and visual displays. Complex functionality is built In to the pilot¶s joy-stick, which has up to 17 functions, including pitch-roll- yaw controls, plus a rich set of buttons and triggers. Similarly automobile controls may have turn signals, wiper settings, and washer buttons all built onto a single stick, and typical video camera controls may have dozens of settings that are adjustable through knobs and switches. Rich designs for hand input can inform users and free their minds for status monitoring and problem solving. The interfering effects of acoustic processing are a limiting factor for designers of speech recognition, but the the role of emotive prosody raises further con-cerns. The human voice has evolved remarkably well to support human-human interaction. We admire and are inspired by passionate speeches. We are moved by grief-choked eulogies and touched by a child¶s calls as we leave for work. A military commander may bark commands at troops, but there is as much motivational force in the tone as there is information in the words. Loudly barking commands at a computer is not likely to force it to shorten its response time or retract a dialogue box. 32 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition Promoters of ³affective´ computing, or reorganizing, responding to, and making emotional displays, may recommend such strategies, though this approach seems misguided. Many users might want shorter response times without having to work them-selves into a mood of impatience. Secondly, the logic of computing requires a user response to a dialogue box independent of the user¶s mood. And thirdly, the uncertainty of machine recognition could undermine the positive effects of user control and interface predictability. 33 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition Chapter 7 APPLICATION One of the main benefits of speech recognition system is that it lets user do other works simultaneously. The user can concentrate on observation and manual operations, and still control the machinery by voice input commands. Consider a material-handling plant where a number of conveyors are employed to transport various grades of materials to different destinations. Nowadays, only one operator is employed to run the plant. He has to keep a watch on various meters, gauges, indication lights, analyzers, overload devices, etc from the central control panel. If something wrong happens, he has to run to physically push the µstop¶ button. How convenient it would be if a conveyor or a number of conveyors are stopped automatically by simply saying stop. Another major application of speech processing is in military operations. Voice control of weapons is an example. With reliable speech recognition equipment, pilots can give commands and information to the computers by simply speaking in to their microphones-they don¶t have to use their hands for this purpose .Another good example is a radiologist scanning hundreds of Xrays, ultra sonograms, CT scans and simultaneously dictating conclusion to a speech recognition system connected to word processors. The radiologist can focus his attention on the images rather than writing the text. Voice recognition could also be used on computers for making airline and hotel reservations. A user requires simply to state his needs, to make reservation, cancel a reservation, or make enquiries about schedule. sitive effects of user control and interface predictability. 7.1 Health care In the health care domain, even in the wake of improving speech recognition technologies, medical transcriptionists (MTs) have not yet become obsolete. Many experts in the field anticipate that with increased use of speech recognition technology, the services provided may be redistributed rather than replaced. Speech recognition is used to enable deaf people to understand the spoken word via speech to text conversion, which is very helpful. Speech recognition can be implemented in front-end or back-end of the medical documentation process. 34 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition y Front-End SR is where the provider dictates into a speech-recognition engine, the recognized words are displayed right after they are spoken, and the dictator is responsible for editing and signing off on the document. It never goes through an MT/editor. y Back-End SR or Deferred SR is where the provider dictates into a digital dictation system, and the voice is routed through a speech-recognition machine and the recognized draft document is routed along with the original voice file to the MT/editor, who edits the draft and finalizes the report. Deferred SR is being widely used in the industry currently. Many Electronic Medical Records (EMR) applications can be more effective and may be performed more easily when deployed in conjunction with a speech-recognition engine. Searches, queries, and form filling may all be faster to perform by voice than by using a keyboard. 7.2 Military 7.2.1 High-performance fighter aircraft Substantial efforts have been devoted in the last decade to the test and evaluation of speech recognition in fighter aircraft. Of particular note are the U.S. program in speech recognition for the Advanced Fighter Technology Integration (AFTI)/F-16 aircraft (F-16 VISTA), the program in France on installing speech recognition systems on Mirage aircraft, and programs in the UK dealing with a variety of aircraft platforms. In these programs, speech recognizers have been operated successfully in fighter aircraft with applications including: setting radio frequencies, commanding an autopilot system, setting steer-point coordinates and weapons release parameters, and controlling flight displays. Generally, only very limited, constrained vocabularies have been used successfully, and a major effort has been devoted to integration of the speech recognizer with the avionics system. Some important conclusions from the work were as follows: 1. Speech recognition has definite potential for reducing pilot workload, but this potential was not realized consistently. 35 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition 2. Achievement of very high recognition accuracy (95% or more) was the most critical factor for making the speech recognition system useful with lower recognition rates, pilots would not use the system. 3. More natural vocabulary and grammar, and shorter training times would be useful, but only if very high recognition rates could be maintained. Laboratory research in robust speech recognition for military environments has produced promising results which, if extendable to the cockpit, should improve the utility of speech recognition in high-performance aircraft. Working with Swedish pilots flying in the JAS-39 Gripen cockpit, Englund (2004) found recognition deteriorated with increasing G-loads. It was also concluded that adaptation greatly improved the results in all cases and introducing models for breathing was shown to improve recognition scores significantly. Contrary to what might be expected, no effects of the broken English of the speakers were found. It was evident that spontaneous speech caused problems for the recognizer, as could be expected. A restricted vocabulary, and above all, a proper syntax, could thus be expected to improve recognition accuracy substantially. The Eurofighter Typhoon currently in service with the UK RAF employs a speakerdependent system, i.e. it requires each pilot to create a template. The system is not used for any safety critical or weapon critical tasks, such as weapon release or lowering of the undercarriage, but is used for a wide range of other cockpit functions. Voice commands are confirmed by visual and/or aural feedback. The system is seen as a major design feature in the reduction of pilot workload, and even allows the pilot to assign targets to himself with two simple voice commands or to any of his wingmen with only five commands. 7.2.2 Helicopters The problems of achieving high recognition accuracy under stress and noise pertain strongly to the helicopter environment as well as to the fighter environment. The acoustic noise problem is actually more severe in the helicopter environment, not only because of the high noise levels but also because the helicopter pilot generally does not wear a facemask, which would reduce acoustic noise in the microphone. Substantial test and evaluation programs have been carried out in the past decade in speech recognition systems applications in helicopters, notably 36 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition by the U.S. Army Avionics Research and Development Activity (AVRADA) and by the Royal Aerospace Establishment (RAE) in the UK. Work in France has included speech recognition in the Puma helicopter. There has also been much useful work in Canada. Results have been encouraging, and voice applications have included: control of communication radios; setting of navigation systems; and control of an automated target handover system. As in fighter applications, the overriding issue for voice in helicopters is the impact on pilot effectiveness. Encouraging results are reported for the AVRADA tests, although these represent only a feasibility demonstration in a test environment. Much remains to be done both in speech recognition and in overall speech recognition technology, in order to consistently achieve performance improvements in operational settings. 7.2.3 Battle management Battle Management command centres generally require rapid access to and control of large, rapidly changing information databases. Commanders and system operators need to query these databases as conveniently as possible, in an eyes-busy environment where much of the information is presented in a display format. Human-machine interaction by voice has the potential to be very useful in these environments. A number of efforts have been undertaken to interface commercially available isolated-word recognizers into battle management environments. In one feasibility study speech recognition equipment was tested in conjunction with an integrated information display for naval battle management applications. Users were very optimistic about the potential of the system, although capabilities were limited. Speech understanding programs sponsored by the Defense Advanced Research Projects Agency (DARPA) in the U.S. has focused on this problem of natural speech interface. Speech recognition efforts have focused on a database of continuous speech recognition (CSR), largevocabulary speech which is designed to be representative of the naval resource management task. Significant advances in the state-of-the-art in CSR have been achieved, and current efforts are focused on integrating speech recognition and natural language processing to allow spoken language interaction with a naval resource management system. 37 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition 7.3 Training air traffic controllers Training for military (or civilian) air traffic controllers (ATC) represents an excellent application for speech recognition systems. Many ATC training systems currently require a person to act as a "pseudo-pilot", engaging in a voice dialog with the trainee controller, which simulates the dialog which the controller would have to conduct with pilots in a real ATC situation. Speech recognition and synthesis techniques offer the potential to eliminate the need for a person to act as pseudo-pilot, thus reducing training and support personnel. Air controller tasks are also characterized by highly structured speech as the primary output of the controller, hence reducing the difficulty of the speech recognition task. The U.S. Naval Training Equipment Center has sponsored a number of developments of prototype ATC trainers using speech recognition. Generally, the recognition accuracy falls short of providing graceful interaction between the trainee and the system. However, the prototype training systems have demonstrated a significant potential for voice interaction in these systems, and in other training applications. The U.S. Navy has sponsored a large-scale effort in ATC training systems, where a commercial speech recognition unit was integrated with a complex training system including displays and scenario creation. Although the recognizer was constrained in vocabulary, one of the goals of the training programs was to teach the controllers to speak in a constrained language, using specific vocabulary specifically designed for the ATC task. Research in France has focused on the application of speech recognition in ATC training systems, directed at issues both in speech recognition and in application of task-domain grammar constraints.[4] The USAF, USMC, US Army, and FAA are currently using ATC simulators with speech recognition from a number of different vendors, including UFA, Inc, and Adacel Systems Inc (ASI). This software uses speech recognition and synthetic speech to enable the trainee to control aircraft and ground vehicles in the simulation without the need for pseudo pilots. Another approach to ATC simulation with speech recognition has been created by Supremis. The Supremis system is not constrained by rigid grammars imposed by the underlying limitations of other recognition strategies. 38 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition 7.4 Telephony and other domains ASR in the field of telephony is now commonplace and in the field of computer gaming and simulation is becoming more widespread. Despite the high level of integration with word processing in general personal computing, however, ASR in the field of document production has not seen the expected increases in use. The improvement of mobile processor speeds made feasible the speech-enabled Symbian and Windows Mobile Smartphones. Speech is used mostly as a part of User Interface, for creating pre-defined or custom speech commands. Leading software vendors in this field are: Microsoft Corporation (Microsoft Voice Command), Nuance Communications (Nuance Voice Control), Vito Technology (VITO Voice2Go), Speereo Software (Speereo Voice Translator) and SVOX. 7.5 People with disabilities People with disabilities can benefit from speech recognition programs. Speech recognition is especially useful for people who have difficulty using their hands, ranging from mild repetitive stress injuries to involved disabilities that preclude using conventional computer input devices. In fact, people who used the keyboard a lot and developed RSI became an urgent early market for speech recognition. Speech recognition is used in deaf telephony, such as voicemail to text, relay services, and captioned telephone. Individuals with learning disabilities who have problems with thought-to-paper communication (essentially they think of an idea but it is processed incorrectly causing it to end up differently on paper) can benefit from the software. 7.6 Hands-free computing Hands-free computing is a term used to describe a configuration of computers so that they can be used by persons without the use of the hands interfacing with commonly used human interface devices such as the mouse and keyboard. Hands-free computing is important because it is useful to both able and disabled users. Speech recognition systems can be trained to recognize specific commands and upon confirmation of correctness instructions can be given to systems without the use of hands. This may be useful while driving or to an inspector or engineer in a 39 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition factory environment. Likewise disabled persons may find hands-free computing important in their everyday lives. Just like visually impaired have found computers useful in their lives. 40 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition 8 . CONCLUSION Speech recognition will revolutionize the way people conduct business over the Web and will, ultimately, differentiate world-class e-businesses. Voice XML ties speech recognition and telephony together and provides the technology with which businesses can develop and deploy voice-enabled Web solutions TODAY! These solutions can greatly expand the accessibility of Web-based self-service transactions to customers who would otherwise not have access, and, at the same time, leverage a business¶ existing Web investments. Speech recognition and Voice XML clearly represent the next wave of the Web. It is important to consider the environment in which the speech system has to work. The grammar used by the speaker and accepted by the system, noise level, noise type, position of the microphone, and speed and manner of the user¶s speech are some factors that may affect the quality of speech recognition. Since, most recognition systems are speaker independent, it is necessary to train a system to recognize the dialect of each user. During training, the computers display a word and the user reads it aloud. 41 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition 9. BIBLOGRAPHY 1. Luger, George; Stubblefield, William (2004). Artificial Intelligence: Structures and Strategies for Complex Problem Solving (5th ed.). The Benjamin/Cummings Publishing Company, Inc.. ISBN 0-8053-4780-1. http://www.cs.unm.edu/~luger/ai-final/tocfull.html. 2. Nilsson, Nils (1998). Artificial Intelligence: A New Synthesis. Morgan Kaufmann Publishers. ISBN 978-1-55860-467-4. 3. Russell, Stuart J.; Norvig, Peter (2003), Artificial Intelligence: A Modern Approach (2nd ed.), Upper Saddle River, New Jersey: Prentice Hall, ISBN 0-13790395-2, http://aima.cs.berkeley.edu/ 4. Poole, David; Mackworth, Alan; Goebel, Randy (1998). Computational Intelligence: A Logical Approach. New York: Oxford University Press. http://www.cs.ubc.ca/spider/poole/ci.html. 5. Winston, Patrick Henry (1984). Artificial Intelligence. Reading, Massachusetts: Addison-Wesley. ISBN 0201082594. 6. www.seminaron.com 7. www.scribd.com 8. www.seminaronly.com 9. http://www.becta.org.uk 10. http://www.edc.org 11. http://www.dyslexic.com 12. http://www.ibm.com 13. http://www.dragonsys.com 14. http://www.out-loud.com 42 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition 43 Dept. of E&C H.K.B.K.C.E. Artificial intelligence for speech recognition 44 Dept. of E&C H.K.B.K.C.E.