There is a relatively large body of research that is aimed at finding a set of acoustic measures of voice signals that can be used to: (a) aid in the detection, diagnosis, and evaluation of voice-q...

A Methodological Study of Perturbation and Additive Noise in Synthetically Generated Voice Signals

Recent advances in automatic speech recognition

A functioning system is described in which spoken Fortran-like programs can be reliably interpreted by a computer. Speech is encoded into cepstral patterns using a shifting overlapping 32-ms Hamming window. A vocabulary of 26 words was selected so that the distance between different words, in their cepstral representation, was large. The syntax of individual phoneme strings in a word, and the syntax chosen for the programming language, were used to resolve ambiguous decisions. Feedback to the programmer greatly increased the reliability of the system, since, 1) mistaken decisions could be corrected, and 2) the programmer gradually learned to speak in such a way that system mistakes were greatly reduced. With two trained speakers, the recognition rate was 50-75 percent on sentences, 75-90 percent on words, and with user feedback and correction, the recognition rate was 100 percent with no more than two repetitions for long statements. The good performance achieved was due to the well-chosen vocabulary, strong syntactic support, and speaker training. The basic drawback for the system at present is that the initial training of the user is a time-consuming process. However, further improvement has since been achieved by using the initial isolated cepstral prototypes to locate new prototypes in samples of continuous speech, and then using the "continuous prototypes" for recognition. Currently, formant trajectories, derived from a pitch synchronous, linear prediction analysis, are being used, and an automatic stress analysis program provides segmentation, and guides the selection of key allophones.

A computer programming system using continuous speech input

There is a relatively large body of research that is aimed at finding a set of acoustic measures of voice signals that can be used to: (a) aid in the detection, diagnosis, and evaluation of voice-quality disorders; (b) identify individual speakers by their voice characteristics; or (c) improve methods of voice synthesis. Three acoustic parameters that have received a relatively large share of attention, especially in the voice-disorders literature, are pitch perturbation, amplitude perturbation, and additive noise. The present stndy consisted of a series of simulations using a general-purpose formant synthesizer that were designed primarily to determine whether these three parameters could be measured independent of one another. Results suggested that changes in any single dhnension can affect measured values of all three parameters. For example, adding noise to a voice signal resulted not only in a change in measured signal-to-noise ratio, but also in measured values ofpftch and amplitude perturbation, These interactions were quite large in some cases, especially in view of the fact that the perturbatio n phenomena that are being measured are generally quite small. For the most part, the interactions appear to be readily explainable when the measurement techniques are viewed in relation to what is known about the acoustics of voice production. Developments in signal-processing technology have made numerous techniques available to the voice scien- tist for analyzing the acoustic characteristics Of human voices. Measurements of vocal characteristics have been applied to a wide range of practical problems. These include research aimed at both identifying individual speakers by their voice characteristics, and determining what acoustic characteristics of voices need to be trans- mitted in low bit-rate speech synthesis systems. There is also an extensive body of research directed toward the development of quantitative measures of the acoustic characteristics associated with laryngeal pathology. The present paper focuses on measurement problems associated with three acoustic parameters that have re- ceived a relative!y large share of attention in the voice- disorders literature: additive noise, pitch perturbation, and amplitude perturbation. The term additive noise is generally used to refer to the acoustic byproduct of turbulent air flow generated at the glottis during phona- tion. Additive noise is often represented on a decibel scale as a ratio of the amount of energy in the periodic or "harmonic" component to the amount of aperiodic en- ergy. Pitch perturbation or "vocal jitter" refers to rapid, and generally relatively small, eycle-to-cye!e variations in the fundamental period of the glottal source function. Amplitude perturbation, or "vocal shimmer" refers to analogous cycle-to-cycle variations in voice amplitude. Although it is seldom stated explicitly, it is generally assumed that measured values of pitch perturbation, amplitude perturbation, and additive noise reflect funda- mentally different sources of aperiodicity in the laryngeal vibratory pattern. It has also been assumed that these three parameters can be measured independent of one another in the voice waveform. The primary purpose of the present study was to use a series of eomputer simu- lations to determine whether pitch perturbation, ampli- rude perturbation, and additive noice can be measured independent of One another.

A methodological s tudy of perturbation a nd additive n oise in synthetically g enerated v oice signals

This research has resulted in the development of a new pitch-synchronous analysis technique for the extraction of accurate formant information from speech signals. The method is an improvement over current methods of analysis in terms of accuracy and temporal resolution. This is achieved by extension of the signal from one pitch period into the next, using a speech production model based on linear prediction. The result is higher accuracy in the determination of formant frequencies, bandwidths and amplitudes, and the ability to follow rapid formant transitions. The method performs equally well with nasal and high pitched sounds. The method is applied to the speech recognition and the speaker identification problems.

Y. Gulut

Papers

A new method for accurate analysis of voiced speech