Phonemic Symbols Research Articles

A speech labeling system based on knowledge processing

The segmentation of continuous speech into phonemes, i.e., labeling, is one of the most important procedures in constructing the speech database. At the present stage, the labeling is executed by the inspection of the time-series of acoustic parameters by the human expert. Because of this procedure, a large amount of time and effort are required in the construction of the speech database. From such a viewpoint, the authors have developed an automatic speech labeling system aiming at the automatic labeling. In the proposed system, a correspondence is established between the phoneme symbol sequence and the characteristic changes in the time-course of the acoustic parameters, and the rough position of the phonemes are determined. Then the boundaries between the phonemes are adjusted by the detailed observation of the change of acoustic parameters near the phoneme boundaries. Finally, the change of the acoustic parameters in each phoneme interval is examined to verify the match of the label. The system is evaluated by experiment. Eight sentences uttered by seven adult males, who participated in the development of rules, are used for evaluation; 6.7 percent of the phoneme boundaries are rejected. For the remaining phoneme boundaries, 99.1 percent of the boundaries are set within 30 ms of the position decided previously by the human expert. For 10 sentences uttered by four adult males, who did not participate in the rule development, the foregoing scores turned out to be 8.9 and 95.5 percent, respectively. Those results indicate the usefulness of the proposed system.

Speech synthesis system of rule-synthesis type

In order to generate a series of speech parameters from a series of phonemic symbols extracted from a series of input characters, parameters for given syllables or phonemes are read from corresponding parameter files according to the types of immediately preceding vowels or consonants of the given syllables or phonemes in the series of phonemic symbols. The syllable or phoneme parameters ar combined to produce a series of speech parameters.

Learning and production of speech pattern using multilayer neural networks

AbstractA new neural network and its learning algorithm are proposed. The neural network consists of four layers—input, hidden, output and final output layers. The hidden and output layers are multiple. The proposed learning algorithm is called the SICL (spread pattern information and cooperative learning) method and has the following features: (1) the singular points of back propagation errors in the BP method are removed; (2) a spread pattern information (SI) algorithm is proposed; and (3) a cooperative learning (CL) algorithm is proposed.Using the SICL method, it is possible to learn analog data accurately and to obtain a stable output. Using this neural network, the authors developed a speech production system consisting of a phonemic symbol production subsystem and a speech parameter production subsystem. The system is applied to speech data and the system performance is examined. Especially, for the speech parameter production subsystem, the learning capacity and the optimal area for learning constants are studied. As a result, it is shown that it is possible to learn and produce speech data with high accuracy using the proposed neural network.

Poetry I: Teaching verse with CAI

The teaching of literature through CAI raises problems of both a linguistic and instructional nature; student involvement and creativity in studying literature, and especially poetry, is difficult to build into a computer-based lesson. We have confronted these difficulties in the lessonPoetry I, which introduces undergraduates to basic concepts of poetic verse in a design using screen display, speech synthesis, and verse processing to maximize interactivity and student involvement. The lesson contains instructional modules which include the student's composition of a limerick. Computational processing of the limerick's text enables the program to offer guidance as the student composes and revises the verse and gains first-hand experience with metrical language. The significant problems inherent in processing verse are addressed by adapting Digital Equipment Corporation's DECTalk speech synthesizer as an engine converting natural language verse text into accessible strings of phonemic symbols. Although the verse processor cannot “scan” verse, it can interpret DECTalk's symbols reliably enough to successfully stimulate students' thinking about their verse compositions. Preliminary responses to the program have been favorable and demonstrate its effectiveness in involving students more deeply in learning about poetic verse. This suggests that the techniques of verse processing prototyped inPoetry I might be usefully extended to other types of verse and levels of study.

Speaker-independent recognition of the DARPA Naval Resource Management Task

In this talk, the issue of acoustic modeling of the fundamental units for recognition is examined. Modeling approaches for both context independent (CI) and context dependent (CD) units are studied. The acoustic modeling approaches were tested on speaker independent recognition of the DARPA Naval Resource Management Task. The set of context independent units in this study is a fixed set of 47 phonelike units (PLUs), in which each PLU is associated with a linguistically defined phoneme symbol. Each CI/PLU is modeled using a continuous density hidden Markov model (CDHMM) with Gaussian mixture state observation density. The set of context dependent units includes PLUs defined by left, right, and both left and right context. Only those CD/PLUs with enough occurrences in the training data are selected for modeling. Two approaches are presented to model the CD/PLUs. Both CI/PLUs and CD/PLUs are obtained using the segmental k-means training procedure. In the case of context independent acoustic modeling, the maximum number of mixtures was varied in each state from 1 to 256 and it was found that the word accuracy increased from 56% to 89% which indicates that sufficient acoustic resolution is essential for improved performance. The 89% word accuracy is the highest performance reported based on context independent units. When context dependency modeling is incorporated, both modeling approaches achieved better than 92% word accuracy.

Keys to phonemic symbols: English phonology on the computer

Keys to phonemic symbols: English phonology on the computer

Learning to Read and Spell Words

Learning to read and spell words is a central part of becoming literate. During text reading, most words are processed, and skilled readers are able to do this effortlessly. How they become skilled at processing graphic cues has been the focus of our research. Findings indicate that prereaders do not acquire graphic skill by learning to read signs and labels in their environment. Rather, mastery of letters is required. Whereas prereaders use visual or context cues to identify words, as soon as children move into reading they shift to letter-sound cues. Initially, words are read by accessing remembered associations between a few letters in spellings and sounds in pronunciations. Later, when decoding skill matures, complete spellings are analyzed as phonemic symbols for pronunciations and are stored in memory. Various studies indicate that having a visual picture of speech in memory is an important part of a person's information-processing equipment. Spellings may influence how words are pronounced, what sounds people think are in words, how quickly people judge spoken word rhymes, how rapidly pronunciations change over time.

Syntactic Decision Rules for Recognition of Spoken Words and Phrases Using a Stochastic Automaton

This study deals with the design of a syntactic decision rule for recognizing an unknown utterance from a set X. The decision rule is expressed as a function of the character string (CS) derived from the test utterance. To obtain the CS, the waveform of the utterance is divided into a large number of frames of roughly equal duration numbered 1, 2,...,n. The ith symbol in the CS is the phonemic symbol obtained by subjecting the ith frame of the waveform to a relatively simple phoneme decision rule, the number of symbols in the CS being n. All the available nonacoustic information such as the lexicon of words in the set X, the possibility of confusion between different phonemes as seen by the phoneme decision rule, etc. is used in the design of the decision rule. The syntactic decision rule can be implemented by a stochastic finite state automaton involving limited memory and computation. The decision rule can also be interpreted as yielding the phrase x which minimizes a distance measure D(x, z) between the phrase x X and the observed CS z. We wili compare this approach with the other approaches such as the Viterbi methods, the distance approaches involving various types of distances, etc.

A spoken word recognition system for unspecified male speakers

A spoken word recognition system for unspecified male speakers is outlined. The system operates in the following four stages. (I) Seven acoustic parameters are extracted every 10 ms from the outputs of the filter bank. The parameters are the frequencies of three spectral local peaks, the speech power and three parameters expressing the gross pattern of the spectrum. (II) Segmentation and the phoneme recognition are carried out. (III) Errors in the segmentation and phoneme recognition are corrected by means of phoneme connecting rules. (IV) The recognized sequence is determined to be the item of the dictionary having a maximum similarity to the recognized phonemic sequence. The similarity between the item of the dictionary and the recognized phonemic sequence is computed using confusion matrices made from the recognition of 17 040 phonemes. Every item of the dictionary is written in phonemic symbols derived from the word in Japanese “kana” letters by simple rules, so that the contents of the dictionary can be easily changed. The scores for word recognition were found to be 84.0% for 166 words uttered by 25 male speakers and 95.0% for 51 words selected from these 166.

The devoicing of voiced fricatives

Devoicing of voiced fricatives is not adequately characterized by a single diacritic on phoneme symbols in different contexts. It is a continuous variable to which (1) place of articulation; (2) position; (3) manner class and voicing of adjacent phones; (4) emphatic stress and (5) free variation in force of articulation can contribute. When it occurs the glottis is at least partially open whatever the place of articulation, but it is not established how much of this opening is cause and how much effect.

Processing Strings of Phonemic Symbols by use of Chinese Characters as a Unit

Processing Strings of Phonemic Symbols by use of Chinese Characters as a Unit

On Computer Transcription of Phonemic Symbols

A possible standard for computer transcription of phonemic symbols is proposed.

Relation of Descriptive Phonetics to Research in Speech Communication

This paper does not report experimental results but is intended to consider some of the more general objectives of research on speech communication. The conversion of physiological and acoustical measurements on speech to discrete units is one of the fundamental purposes of speech research. There is a substantial number of physiological parameters and of values within these parameters which are significant in the various languages of the earth. Descriptive phonetics provides a discrete symbolization for the numerous sounds which may be produced by the human vocal mechanism and which are significant in spoken language. An unambiguous and relatively complete system of phonetic notation is essential to experimental research on speech communications. The symbols may be specified by the phonetic parameters and parameter values, and each symbol should denote a unique set of phonetic parameter values. This is not true of phonemic symbols, however, since different tokens of the same phonetic type may serve as members of allophones of different phonemes.

Speech Synthesis by Rule Controlled by a Small, Low-Speed Digital Computer

Computer synthesis of speech by rule has already been demonstrated by workers at the Bell Telephone Laboratories, using a large high-speed computer to perform the entire process. This paper describes a way of achieving a similar result by the use of a small computer, in conjuction with a moving-resonance speech synthesizer. The computer program is fairly simple, and consists of instructions for interpreting a table of phoneme data that has been previously fed into the computer store. The sequence of phonemes corresponding to the speech required, together with associated larynx frequencies, is then fed in from a punched tape. Durations are not specified separately, but are calculated from the tables; however, if variations of duration, intensity, or other properties are required to produce variations of stress, etc., additional characters may be added immediately prior to any phoneme symbol. These characters cause the stored table of properties to be modified for that one occurrence of the associated phoneme. This use of modifying characters enables the program to be easily used for experimental changes of the synthesis rules. These are needed mainly for perception experiments leading to further improvements of the rules.

Features of Coarticulation in VCV Utterances

Spectrographic and cinéradiographic data on voiced stops in V-V context are presented. The data suggest that (1) the neural signals controlling the activity of the vocal structures during V-stop-V utterances do not occur as a discrete time sequence corresponding to the graphic sequence of phoneme symbols; (2) the articulatory response to the neural instruction for the stop is superimposed on the initial portion of the response to the instruction for the following vowel; (3) the vocal-tract-shape target for the stop is not unique and depends on the V-V context as a whole; and (4) the formant loci of a stop+vowel (vowel+stop) syllable may be shifted along the frequency scale by a preceding (following) vowel. Interpretation of the data in relation to a dynamical model of speech production are discussed.

Delimitation of Syntactic Units

In the present state of our techniques we may assume that we know how to isolate morphemes properly-that is, unequivocally and without unaccountable residue. It is not so certain that we know how to isolate words, and hence how to separate morphology from syntax. Given a viable definition of the word for some particular language, we come finally to the problem of assigning words properly to relevant syntactic units such as phrases, clauses, and sentences. To illustrate one possible technique of discovering the syntactic framework of a language, I shall analyze a short text of Ponapean, a Micronesian language of the Eastern Caroline Islands.1 The text is an excerpt from a spontaneous conversation recorded on tape and later transcribed. Roman numerals mark four UTTERANCE UNITS-stretches of speech activity by a single speaker preceded and followed by silence.2 Each utterance unit consists of one or more CONTOURS (marked by slant lines), defined as a phoneme sequence between two audible pauses or between pause and silence, such that all morphophonemic processes take place within the contour and only within this. Every contour has one primary stress and a pitch sequence; it may also have secondary stresses. (There are differences among Ponapean contours which affect the classification of these segments but not the segmentation itself. They will be disregarded in this paper.) In the text, numbers in parentheses identify the individual WORDS (but hesitation noises are not numbered). The word is defined, for Ponapean, as a sequence of morphemes and morpheme clusters in fixed order. Sequences with variable order consist of separate one-morpheme words. The following phoneme symbols appear in the text: SHORT VOWELS a o e i, D;