D. Dwayne Paschall

Track-draw: A graphical interface for controlling the parameters of a speech synthesizer

In this report we describe a graphical interface for generating voiced speech using a frequency-domain implementation of the Klatt (1980) cascade formant synthesizer. The input to the synthesizer is a set of parameter vectors, calledtracks, which specify the overall amplitude, fundamental frequency, formant frequencies, and formant bandwidths at specified time intervals. Tracks are drawn with the aid of a computer mouse that can be used either inpoint-draw mode, which selects a parameter value for a single time frame, or inline-draw mode, which uses piecewise linear interpolation to connect two user-selected endpoints. Three versions of the program are described: (1) SYNTH draws tracks on an empty time-frequency grid, (2) SPECSYNTH creates a spectrogram of a recorded signal upon which tracks can be superimposed, and (3) SWSYNTH is similar to SPECSYNTH, except that it generatessine-wave speech (Remez, Rubin, Pisoni, & Carrell, 1981) using a set of time-varying sinusoids rather than cascaded formants. The program is written for MATLAB, an interactive computing environment for matrix computation. Track-Draw provides a useful tool for investigating the perceptually salient properties of voiced speech and other sounds.

Demystifying incomplete neutralisation during final devoicing

Phonetic incompleteness during final devoicing has been subject to some debate, particularly as far as the reliability of the data and the proper explanation of incompleteness are concerned. This paper approaches final devoicing and incompleteness from the perspective of the functions they perform in the phonological system. Production and perception data on final devoicing in Afrikaans are presented. The data indicate that under more natural speaking conditions, final devoicing is complete, but various reflexes of incomplete neutralisation are observed under different types of experimental conditions. When they are present, some of these reflexes do play a role in the perception of the difference between word pairs ending on underlying voiced and voiceless plosives. The interpretation of the data proposed in the paper is that final devoicing itself is a complete and categorical process affecting the feature [voice], aimed at maintaining the perceptual salience of final obstruents generally. However, when pragmatic factors should demand the maintenance of the otherwise neutralised opposition, Afrikaans speakers implement a phonetic difference along the dimension of the complementary feature [tense], leading to incomplete phonemic neutralisation, to perform the function of disambiguation.

Identification of high-pass filtered male, female, and child vowels: The use of high-frequency cues

Vowels are characteristically described according to low-frequency resonance characteristics, which are presumed to provide the requisite information for identification. Classically, the study of vowel perception has focused on the lowest formant frequencies, typically F1, F2, and F3. Lehiste and Peterson [Phonetica 4, 161-177 (1959)] investigated identification accuracy of naturally produced male vowels composed of various amounts of low- and high-frequency content. Results showed near-chance identification performance for vowel segments containing only spectral information above 3.5 kHz. The authors concluded that high-frequency information was of minor importance for vowel identification. The current experiments report identification accuracy for high-pass filtered vowels produced by two male, two female, and two child talkers using both between- and within-subject designs. Identification performance was found to be significantly above chance for the majority of vowels even after high-pass filtering to remove spectral content below 3.0-3.5 kHz. Additionally, the filtered vowels having the highest fundamental frequency (child talkers) often had the highest identification accuracy scores. Linear discriminant function analysis mirrored perceptual performance when using spectral peak information between 3 and 12 kHz.

Ranking the pitches of concurrent vowels

When two vowels are presented simultaneously, listeners can label them more accurately if their fundamental frequencies differ, and under some conditions they can report which vowel has the higher pitch. In this study the interaction of fundamental frequency and binaural cues on judgments of the relative pitches of concurrent vowel pairs is examined. On each trial listeners were told which vowels were presented and were asked to indicate which vowel was on the higher pitch. The first vowel had an F0 of 140 Hz and the second vowel was either 1, 2, 4, or 8 semitones lower or higher than the first. When the fundamental frequency difference was small (1 semitone), the ability to rank the pitches of the two vowels was near chance level, and dichotic presentation provided little advantage over monaural presentation. With monaural presentation performance improved gradually with increasing F0 difference, reaching a maximum near 80% correct at +8 semitones. When the stimuli were presented dichotically there was a...

Enhancement of concurrent vowels: Postcursor effects and thresholds.

Speech communication often takes place in noisy environments, including situations where two or more voices compete for the attention of the listener. Experiments with simultaneous vowel pairs investigated aspects of a priming stimulus that draws attention to a target voice in a mixture of two voices. Previous studies have shown identification accuracy is higher in conditions where the vowel pair was preceded by a precursor with the same F0 and spectrum envelope as one of the vowels. These findings are consistent with an explanation based on auditory adaptation. This conclusion is reinforced by results from the present experiments which show that when the precursor is placed after the concurrent vowel pair (a postcursor), there is little benefit to identification accuracy. Thresholds for identifying a target vowel in the presence of a masker vowel when a precursor is present with the same intensity, F0, and spectrum envelope as the masker vowel suggests that the phenomenon of enhancement appears consisten...

Identification of talker gender and speech production mode from high-pass filtered vowel segments

This study reports gender and production mode judgments from high-pass filtered vowel segments from 22 listeners with normal-hearing. Two adult males and two adult females produced vowels in an h/Vowel/d (hVd) context in both a spoken and sung mode of production. The hVd utterances were produced using the carrier phrase, “I say (insert hVd) again.” The speakers produced the sung utterances to the tune of “Old McDonald.” The vowels included /i/ as in “heed”, /ɝ/ as in “heard”, and /ɔ/ as in “hawd.” A 250 millisecond segment from the central portion was extracted and windowed. The signals were then high-pass filtered at 3.2 kHz to remove low-frequency spectral detail. The listeners were positioned in front of a graphical user interface and instructed to click on one of four buttons, “Male spoken,” “Male sung,” “Female spoken,” or “Female sung” to indicate the perceived gender and mode of production. Results showed below chance performance for the male sung and female spoken utterances and performance signif...

The identification of high pass filtered vowels

Vowels are typically described according to their spectral prominences (i.e., formants). Previous studies have shown that the first three formants provide important information for identification (Hillebrand et al. (1995), Miller (1989), Molis (2005), Peterson and Barney (1952)]. The present study measured identification accuracy for six naturally produced vowels spoken by a male and female talker with these spectral prominences removed. The six hVd tokens for each talker were high-pass filtered to remove the first three formants from the vowels and then identified by 24 normal hearing listeners. Results suggest that listeners identified a majority of the tokens above chance levels. The average identification of the male vowels was 29% (range, 17%–47%), with two vowels identified with nearly 50% accuracy. Average identification for the six female vowels was 53% (range, 37%–72%), with 5 of 6 vowels being identified with over 40% accuracy.

Neural‐network‐based voice‐tracking algorithm

A voice‐tracking algorithm was developed and tested for the purposes of electronically separating the voice signals of simultaneous talkers. Many individuals suffer from hearing disorders that often inhibit their ability to focus on a single speaker in a multiple speaker environment (the cocktail party effect). Digital hearing aid technology makes it possible to implement complex algorithms for speech processing in both the time and frequency domains. In this work, an average magnitude difference function (AMDF) was performed on mixed voice signals in order to determine the fundamental frequencies present in the signals. A time prediction neural network was trained to recognize normal human voice inflection patterns, including rising, falling, rising–falling, and falling–rising patterns. The neural network was designed to track the fundamental frequency of a single talker based on the training procedure. The output of the neural network can be used to design an active filter for speaker segregation. Tests...

Binaural enhancement of simultaneous vowels

Previous investigations have shown that for monaural presentations, identification accuracy of simultaneous vowels improves when a precursor with the same F0 and spectrum envelope is present which draws attention to a particular voice. In normal listening situations, however, one likely makes use of both ears rather than one ear only. The present investigation examined the ability of a dichotic precursor to enhance simultaneous vowel sounds presented to the left ear. A precursor was presented to the listeners whose intensity level in the left ear was the same as that used in previous investigations. However, the intensity of the precursor in the right ear was varied to produce different perceptual locations of the precursor: left ear only, 45° left of midline, midline, 45° right of midline, and right ear only. Thus the same monaural stimulus that has been shown to produce enhancement in previous investigations was present in each precursor location. However, the perceptual location of the precursor was di...

Modeling vowel identification by listeners: A comparison of cepstrum‐based and auditory models at reduced signal‐to‐noise ratios

This study compared the abilities of two computational models to predict the identification responses of listeners to vowels presented in quiet and with pink‐noise maskers at signal‐to‐noise ratios (SNRs) of −6, −3, 0, +3, and +6 dB. One model used a representation derived from cepstral analysis. The other used a representation derived from a model of peripheral auditory analysis which (i) filtered the stimulus with a bank of bandpass filters; (ii) compressed the filtered waveforms using a simulation of hair cell transduction; (iii) computed the short‐term autocorrelation function of the compressed, filtered waveform in each channel; and (iv) summed the autocorrelation functions across filter channels to generate a pooled autocorrelation function (PACF). Classification was performed by discriminant function analysis, using noise‐free tokens as input to the classifier during the training phase. Performance using the cepstrum‐based model was better than human listeners in quiet, but deteriorated rapidly wit...