Speech Of Talkers Research Articles

Effects of Speech-Driven Embodied Entrainment Character "InterActor" on Talker's Speech

A speech-driven embodied entrainment character called “InterActor” with both functions of speaker and listener has been developed for activating human interaction by generating the whole body motions such as nodding, blinking and the actions of head, arms and waist coherently related to speech input. This paper demonstrates the effects of InterActor on interaction in human speech input by the analysis of speech temporal parameters. It is found that speech duration increases and speech activity decreases because of the increase of total silence duration and the number of silence parts in human-to-InterActor speech input, which is the advantage of real time processing for speech recognition and so forth. Interaction with InterActor is also highly evaluated in smoothness, enjoyment and preference by sensory evaluation.

The effect of fear on paralinguistic aspects of speech in patients with panic disorder with agoraphobia

The present study investigated the effect of fear on paralinguistic aspects of speech in patients suffering from panic disorder with agoraphobia ( N = 25). An experiment was conducted that comprised two modules: Autobiographical Talking and Script Talking. Each module consisted of two emotional conditions: Fearful and Happy. Speech was recorded digitally and analyzed using PRAAT, a computer program designed to extract paralinguistic measures from digitally recorded spoken sound. In addition to subjective fear, several speech characteristics were measured as a reflection of psychophysiology: rate of speech, mean pitch and pitch variability. Results show that in Autobiographical Talking speech was slower, had a lower pitch, and a lower pitch variability than in Script Talking. Pitch variability was lower in Fearful than in Happy speech. The findings indicate that paralinguistic aspects of speech, especially pitch variability, are promising measures to gain information about fear processing during the recollection of autobiographical memories.

Controlling the dog, pretending to have a conversation, or just being friendly?

This study examines the effects of sex and familiarity on Americans’ talk to dogs during play, using categories derived from research comparing mothers’ and fathers’ talk to infants. Eight men and fifteen women were videotaped whilst playing with their own dog and with another person’s dog, and their utterances were codified for features common to infant-directed talk. Women used the baby talk speech register more than men, and both men and women used this register more when interacting with the unfamiliar dog than with the familiar dog. When playing with the familiar dog, women talked more than men, and their talk was more suggestive of friendliness and having a conversation. When playing with the unfamiliar dog people used more praise, more conversational gambits, a more diverse vocabulary, and longer utterances than when playing with the familiar dog, suggesting that when playing with the unfamiliar dog, people pretended to have more of a conversation, were more attentive to appearing friendly and were less attentive to the dog’s limited understanding. Overall, however, men and women used similar forms of talk when interacting with a dog, whether familiar or not.

Simultaneous Recognition of Distant-Talking Speech of Multiple Talkers Based on the 3-D N-Best Search Method

This paper describes a novel method for hands-free speech recognition and in particular for simultaneous recognition of distant-talking speech of multiple sound sources (talkers or noise sources). Our method is based on the 3-D Viterbi search extended to a 3-D N-best search method to allow simultaneous speech recognition of multiple talkers. The baseline system integrates two existing technologies—3-D Viterbi search and conventional N-best search—into a complete system. However, initial evaluation of the 3-D N-best search-based system showed that new ideas were needed in order to build a system to simultaneously recognize multiple sound sources. Two factors were found to have an important role in system performance. Those two factors are the different likelihood ranges of the talkers and the direction-based separation of the hypotheses. More specifically, since we have to compare hypotheses originating from different talkers, an accurate comparison of these hypotheses cannot be made due to the different likelihood dynamic range of the talkers. Moreover, the hypotheses originated from talkers are located in different directions and therefore separating them based on their direction provides an efficient method for accurate recognition. To solve these problems, we implemented a likelihood normalization technique and a path distance-based clustering technique into the baseline 3-D N-best search-based system. The performance of our system was evaluated by experiments for recognizing the distant-talking speech of two talkers. The experiments were carried out on simulated (with only time delay) data and on reverberated (simulated and real) data. In this paper, we evaluated the proposed method in reverberant environments, and we introduced results obtained by experiments at several reverberation times and results obtained in a real environment. The experiments showed that implementing the two techniques described above produced significant improvements. Best results for simulated data were obtained by implementing the two techniques and using a microphone array composed of 32 channels. In that case in particular, the Simultaneous Word Accuracy (where both talkers are correctly recognized simultaneously) in the ‘top 1’ hypothesis was 72.49%, and in the ‘top 3’ hypotheses was 86.25%, which were very promising results.

Temporal characteristics of Chinese-accented English: Preliminary findings

This study investigated several temporal features of English to determine the extent of their occurrence in the speech of talkers of Chinese-accented English who had relatively limited experience with spoken English. Specifically, the extent to which these speakers produced the following temporal contrasts was examined: (1) tense versus lax vowel duration, (2) vowel duration before voiced versus voiceless consonants, and (3) vowel and consonant duration in sentence-final versus nonfinal position. Preliminary data from sentences produced by eight non-native and eight native talkers indicates that the native English speakers and the Chinese-accented talkers did not differ in the extent to which they realized the inherent duration difference between tense and lax vowels. However, the native English subjects tended to show substantially greater vowel lengthening before voiced versus before voiceless consonants than the Chinese-accented talkers. In addition, while the two groups did not differ significantly in the extent to which they lengthened sentence-final consonants relative to nonfinal consonants, the native English talkers showed greater sentence-final vowel lengthening than the Chinese-accented talkers. When group differences for a given temporal parameter were found, 1 or 2 of the non-native subjects typically fell within the range of performance shown by the native speakers.

Segmental Timing in Amyotrophic Lateral Sclerosis

This study describes segmental timing patterns for a passage read by speakers with ALS and a group of age- and gender-matched neurologically healthy speakers. Segment durations for ALS speakers' habitual rates (ALS condition) were compared to segment durations for healthy speakers' habitual and slow reading rates (Control Habitual and Control Slow conditions). Temporal differences for pairs of sound classes (i.e., long-short vowels, voiceless-voiced fricatives, voiceless-voiced stop gaps, high-low vowels) and prepausal lengthening of vowels also were studied. The results of the ALS-Control Habitual comparison suggested that the slowed articulatory rate of ALS influences segmental timing for most phonetic events. The finding that local segmental cues for disordered speakers were broadly consistent with their overall (slow) articulatory rate suggests that duration likely does not contribute in a significant way to intelligibility deficits in ALS. A qualitative analysis further indicated that the slowed rate of ALS is not uniformly distributed across phonetic events, but influences segment durations of particular sound categories in ways consistent with the disease process. The analysis comparing segment durations for the ALS and Control Slow conditions suggested that the overall pattern of segmental timing for speakers with ALS was broadly similar to the pattern for voluntarily slowed speech of healthy talkers. Finally, temporal differences for sound classes (i.e., voiceless-voiced fricatives, long-short vowels) and prepausal lengthening for speakers with ALS typically were similar to healthy speakers' habitual and slow speech. There were substantial variability within speaker groups and reversals of expected effects, however. The implication is that temporal differences for sound classes, such as short and long vowels, are not a particularly reliable perceptual cue for listeners.

Mapping the Range of Information Contained in the Iconic Hand Gestures that Accompany Spontaneous Speech

This study tested McNeill’s theory that the iconic gestures that accompany speech in everyday talk convey critical information in interpersonal communication. Using a structured interview to measure the amount of information respondents receive from clause-length clips depicting aspects of a cartoon story, we discovered that when respondents could see the iconic gestures as well as hear the speech they received significantly more accurate information about those aspects of the original story depicted in the clips than when they just heard the speech. We also discovered that it was only with respect to certain semantic categories, namely, the relative position and the size of objects, that the beneficial effect of gestural communication was significant. We then considered in detail how specific attributes of actions and of objects are communicated via the iconic representation within individual gestures. Lastly, we discuss the implications of these findings for models of human communication in this area.

The perceptual consequences of overlap in /s/ and /∫/ productions within a talker

A primary issue in speech perception is the apparent lack of invariance between the acoustic information in a signal and the listeners perception. Different intended phonemes may be produced with identical acoustic values. Previously, we examined fricative centroids and frication peaks for over 100 utterances beginning with /s/ and /∫/ from each of 20 different speakers, and found substantial overlap across talkers. In the present study, we examine the effect of this overlap on perception. Listeners in a phoneme identification task heard natural productions of /s/ and /∫/ syllables from speakers with either little or great overlap between categories. Although labeling accuracy was near ceiling, effects were found in listeners’ reaction times. Listeners needed more time to interpret the speech of talkers who had substantial overlap in their fricative centroids, even when these talkers showed no overlap in their frication peaks. Listeners also had slower reaction times for talkers with overlap in their frication peaks, but not their centroids. This suggests that category overlap has measureable consequences to perception, and that peaks and centroids are sensitive to different aspects of the speech signal, both of which are perceptually impotant to listeners. [Work supported by NIDCD Grant R01-DC00219 to SUNY at Buffalo.]

Monolingual children in a bilingual situation: Protest, accommodation, and linguistic creativity

The homestay is a common second-language learning situation: non-native speakers live with families of native speakers in hopes of improving knowledge of both language and culture. This paper explores language use in a homestay in which host family members know or are learning the language of the visitor. Accommodation in language choice (Giles et al. 1973; Giles et al. 1991) is one issue in this situation: we explore the extent to which guest and hosts attempt to converge to the other's language. While there is not much convergence by the host family to the guest's language, there is nonetheless significant accommodation to the guest in re-using her linguistic creations in the host language such as jokes and idiom translations. Thus, native speaker and non-native speaker collaborate in the joint construction of an intercultural microcommunity variety (IMV), which differs from interlanguage, foreigner talk and microcommunity speech norms.

Cochannel speaker separation by harmonic enhancement and suppression

This paper presents a system for separating the cochannel speech of two talkers. The proposed harmonic enhancement and suppression (HES) system is based on a frame-by-frame speaker separation algorithm that exploits the pitch estimate of the stronger talker derived from the cochannel signal. The idea behind this approach is to recover the stronger talker's speech by enhancing their harmonic frequencies and formants given a multiresolution pitch estimate. The weaker talker's speech is obtained from the residual signal created when the harmonics and formants of the stronger talker are suppressed. An automatic speaker assignment algorithm is used to place recovered frames from the target and interfering talkers in separate channels. Automatic speaker assignment performs reasonably well in most cochannel environments, including voiced-on-voiced, voiced-on-unvoiced, unvoiced-on-unvoiced, assignment after processing silence intervals, and single talker speech (no cochannel interference). The HES system has been tested at target-to-interferer ratios (TIRs) from -18 to 18 dB with widely available data bases. It has demonstrated improved performance in keyword spotting tests for TIR values of 6, 12, and 18 dB, and in human listening tests for TIR values of -6 and -18 dB.

Hands-free speech recognition by a microphone array and HMM composition

Hands-free speech interface is one of the final goals of human–machine interface. This paper introduces two methods for distant talking speech recognition in noisy and reverberant rooms. The first method is speech recognition using a microphone array. The microphone array enables one to enhance a speech signal using spatial phase differences even in environments where unstationary noises exist. The proposed method is composed of two modules: (1) a high SNR signal retrieval by a delay-and-sum beamformer and (2) localization and trace of the speaker’s direction by extracting a signal power and pitch harmonics. The second method is speech recognition based on the HMM composition. The proposed HMM composition is obtained by extending the HMM composition method of an additive noise to that of the convolutional acoustical transfer function. The HMMs are prepared beforehand for clean speech, noise, and acoustical transfer function. Then the HMM composition is conducted twice for a speech HMM and an acoustical transfer function HMM in the cepstrum domain and for the distorted speech HMMs and noise HMM in a linear spectral domain. The speaker-dependent/independent word recognition experiments using tied-mixture monophone HMMs are carried out and have clarified the effectiveness of the proposed methods. Furthermore, an effective coupling of these methods is also discussed.

A robust algorithm for word boundary detection in the presence of noise

The authors address the problem of automatic word boundary detection in quiet and in the presence of noise. Attention has been given to automatic word boundary detection for both additive noise and noise-induced changes in the talker's speech production (Lombard reflex). After a comparison of several automatic word boundary detection algorithms in different noisy-Lombard conditions, they propose a new algorithm that is robust in the presence of noise. This new algorithm identifies islands of reliability (essentially the portion of speech contained between the first and the last vowel) using time and frequency-based features and then, after a noise classification, applies a noise adaptive procedure to refine the boundaries. It is shown that this new algorithm outperforms the commonly used algorithm developed by Lamel (1981) et al. and several other recently developed methods. They evaluated the average recognition error rate due to word boundary detection in an HMM-based recognition system across several signal-to-noise ratios and noise conditions. The recognition error rate decreased to about 20% compared to an average of approximately 50% obtained with a modified version of the Lamel et al. algorithm.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Talker speech level control circuit for telephone transmitter by piezoelectric conversion

First Page

McGurk effect in non-English listeners: Few visual effects for Japanese subjects hearing Japanese syllables of high auditory intelligibility

The McGurk effect is a phenomenon that demonstrates a perceptual fusion between auditory and visual (lip-read) information in speech perception under the condition of audio-visual discrepancy, created by dubbed video tapes. This paper investigated whether or not the McGurk effect could be extended to Japanese subjects listening to Japanese syllables of different auditory intelligibility. The audio and video signal of a female talker's speech for ten Japanese syllables (/ba/, /pa/, /ma/, /wa/, /da/, /ta/, /na/, /ra/, /ga/, /ka/) was combined on videotapes, giving 100 audio-visual stimuli. These stimuli were presented to ten Japanese subjects who were required to identify the stimuli as heard speech in both noise-added and noise-free conditions. For both conditions, the intelligibility of the auditory stimuli was measured, by presenting the audio-alone stimuli. The results showed that, in the noise-free condition, the McGurk effect was small and almost limited to auditory stimuli of which the intelligibility was less than 100%. In the noise-added condition, the McGurk effect was very strong and widespread. These results indicate that the "Japanese McGurk effect" is less easily induced than the English one, and that it depends on the auditory intelligibility of the speech signal.

Ear Level Recordings of the Long-Term Average Spectrum of Speech*

The long-term average speech spectrum (LTASS) was measured at two different recording positions: 30 cm directly in front of the talker (reference position), and at the tragus of the talker's ear (ear-level position), for three groups of subjects: adult males, adult females, and children. Results indicated significant differences in the overall level and frequency spectra between the LTASS obtained at each microphone location. For all three groups of subjects the LTASS measured at the ear-level position consisted of more low frequency energy (i.e., below 1000 Hz) and less high frequency energy (i.e., above 2500 Hz) than did the LTASS measured at the reference microphone position. The findings suggest that the algorithms currently used to prescribe hearing aid gain may underestimate the sensation level of a hearing-impaired individual's own amplified speech productions at frequencies below 1000 Hz and overestimate the sensation level of a talker's own speech above 2500 Hz. The implications of these findings concerning selection of the electroacoustic characteristics of an amplification system for hearing-impaired individuals are discussed.

Talk and draw: bundling speech and graphics

A description is given of a workstation with integrated graphical and voice I/O (input/output), to improve the man-machine interface (MMI) of the Airborne Warning and Control System (AWACS). Incremental prototype development is used to improve the MMI for AWACS. Operator tasks were analyzed to identify functions that would be better performed with voice I/O, and a 70-word vocabulary and corresponding grammar were developed for an ITT VRS-1280 speech recognizer. Responses for a DEC Talker speech synthesizer were also developed. The system architecture for integrating voice and graphical input in the AWACS interface prototype and its implementation are described and development issues are examined. Future developments are considered.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Evaluation of Four Spanish Word-Recognition-Ability Lists

Tests of word recognition ability (formerly Speech Discrimination) should be suitable for the linguistic background of the person being tested. Few efforts have been made to develop a standardized test in the Spanish language. Current literature reports on six studies that have several drawbacks. The purpose of the present study was to evaluate a commercially available Spanish language test. The material consists of four lists of 50 bisyllabic tetraphonemic Spanish words. The words were recorded at a professional laboratory (Auditec of St. Louis) by a native Spanish speaker. In the present study, the lists were evaluated in terms of interlist equivalence, word difficulty, intelligibility of the talker, and slope of the performance/intensity (PI) function. Taped lists were presented to 16 normal-hearing native Spanish speaking adults at four presentation levels. Mean intelligibility scores were poorest for list three. Statistical analysis indicated that the intelligibility of list three is significantly different than the other lists at the 0.05 level. On the average, at low presentation levels, the nine subjects of Mexican origin obtained better scores than the seven subjects of other nationalities. The slope of the PI-PB function (4.3%) was comparable to that obtained by other investigators of English lists. The words most frequently missed contain the /s/ sound and are words that retain their meaning even after deletion of the final /s/. The talker's speech intelligibility was judged to be very clear.

Voice conversion

We describe some experiments in voice-to-voice conversion that use acoustic parameters from the speech of two talkers (source and target). Transformations are performed on the parameters of the source to convert them to match as closely as possible those of the target. The speech of both talkers and that of the transformed talker is synthesized and compared to the original speech. The objective of this research is to develop a model for (1) creating new synthetic voices, (2) studying factors responsible for synthetic voice quality, and (3) determining methods for speaker normalization.

Perceptual normalization of talker differences

Different talkers may produce the same phoneme with different acoustic patterns and different phonemes with similar acoustic patterns. In order to recognize speech correctly, a listener must use information about the talker's vocal characteristics. A speeded target‐monitoring task was used to investigate whether each token of speech contains sufficient information for normalization or whether a sample of a talker's speech must be heard to normalize subsequent segments. Subjects listened to a sequence of syllables for a particular target. In one type of trial, targets and distractors were produced by a single talker; in other trials, targets and distractors were produced by a mix of talkers. If each token of speech is self‐normalizing, there should be no difference between conditions. Otherwise, performance should be worse in mixed‐talker trials. In another experiment, the available cognitive capacity for recognizing speech was manipulated and recognition of speech produced by a single talker compared to a...

Group Vocal Patterns and Leadership in India

This study was concerned with the effects of language and sex of subjection social and task conversations among university students in India. Each of 20 same-sex 5-person groups, 10 male and 10 female, participated in three 10-minute discussions. The social conversation was in English or Telugu and the two task discussions were in English and Telugu, in counterbalanced order. Results showed more talking and simultaneous speech in Telugu than in English discussions. Although leaders talked more than nonleaders, the correlation was lower than that found in the United States. Analyses suggest that, cross-culturally, leadership is more highly correlated with turns, which consist of both talking and silence, rather than with talking alone.