Human Talker Research Articles

Objective validation of alternative speech test sources for person-to-person speech intelligibility assessments

A sound source that mimics the acoustics characteristics of a human talker is required in objective assessments of speech intelligibility in person-to-person speech communications (PP) scenarios. The few special speech test sources available that meet the strict specifications of the relevant standards are costly and can be inaccessible to practitioners. This can prevent investigations or can lead to non-special and more affordable loudspeakers to be used as alternative sources without knowledge of their suitability. This study evaluates the validity of non-special loudspeakers in PP assessments and supplements a previous preliminary examination by consolidating and expanding early findings. Speech Transmission Index Public Address (STIPA) assessments were conducted employing three representative non-special loudspeakers and one standardized reference speech test source under laboratory conditions. Absolute errors of STIPA measurements were analysed and evaluated against two just perceivable difference (JND) threshold levels. The mean absolute errors obtained for all non-special loudspeakers for almost all scenarios were within one JND of any of the two JND levels. The striking and consistent low level of error found validated the proposition that non-special and affordable loudspeakers may be used as suitable speech sources in pilot or survey-grade speech intelligibility assessments in place of a special standardized speech source when that source is not available, or in full investigations when the relevant standard to adhere to allows the use of those alternative speech sources.

Estimating speaker direction on a humanoid robot with binaural acoustic signals.

To achieve human-like behaviour during speech interactions, it is necessary for a humanoid robot to estimate the location of a human talker. Here, we present a method to optimize the parameters used for the direction of arrival (DOA) estimation, while also considering real-time applications for human-robot interaction scenarios. This method is applied to binaural sound source localization framework on a humanoid robotic head. Real data is collected and annotated for this work. Optimizations are performed via a brute force method and a Bayesian model based method, results are validated and discussed, and effects on latency for real-time use are also explored.

Non-Special Loudspeakers as Speech Test Sources in Natural Acoustics Speech Intelligibility Investigations

Objective speech intelligibility estimations undertaken in natural acoustics speech communications (NAS) scenarios require the utilization of a speech source that approximates the acoustic characteristics of a human talker. Only a limited number of special speech sources that conform to the specifications in the relevant guidelines are available in the market; however, they can be deemed expensive by professional practitioners and other users. Non-special and affordable loudspeakers are often used in NAS investigations in place of standardized special speech sources without the knowledge of their suitability and results validity. This study aims to examine the suitability of a range of representative common and affordable non-special loudspeakers as a potential alternative to standardized speech sources in NAS indicative or pilot investigations. Frequency response and Speech Transmission Index Public Address (STIPA) experimental results obtained from a reference standardized speech source were compared against results from various non-special loudspeakers measured utilizing diverse and real-world representative combinations of NAS acoustic conditions under controlled laboratory conditions. STIPA mean absolute errors for the alternative speech sources were generally lower than the STIPA method uncertainty and one Just Noticeable Difference (0.03 STI). The findings of this study will inform practitioners of the suitability of affordable loudspeakers when standardized special test loudspeakers are not available.

The Influence of Face Masks on Verbal Communication in Persian in the Presence of Background Noise in Healthcare Staff.

Wearing face masks has resulted in verbal communication being more challenging during the COVID-19 pandemic. This study aimed to investigate the effect of face masks on the speech comprehensibility of Persian nurses in healthcare settings. Twenty female nurses from the governmental hospitals randomly participated in an experiment on seven typical commercial face masks at two background noise levels. Nurses' speech intelligibility from a human talker when wearing each face mask was determined based on the speech discrimination score. The vocal effort of nurses wearing each face mask was determined based on the Borg CR10 scale. Based on the linear mixed model, the speech intelligibility of nurses from a human speaker wearing surgical masks, N95 masks, and a shield with face masks were approximately 10%, 20%, and 40–50% lower, respectively, than no-mask conditions (p < 0.01). The background noise decreased the speech intelligibility of nurses by approximately 22% (p < 0.01). The use of a face shield further decreased speech intelligibility up to 30% compared to using a face mask alone (p < 0.01). The vocal efforts of nurses when wearing surgical masks were not significant compared with the baseline vocal efforts (p > 0.05); however, vocal efforts of nurses when wearing N95 and N99 respirators were at an unacceptable level. The face masks had no considerable effect on the speech spectrum below 2.5 kHz; however, they reduced high frequencies by different values. Wearing face masks has a considerable impact on the verbal communication of nurses in Persian. The level of background noise in the healthcare setting can aggravate the effect sizes of face masks on speech comprehensibility.

Comparison of the Acoustic Effects of Face Masks on Speech

Comparison of the Acoustic Effects of Face Masks on Speech

Acoustic effects of medical, cloth, and transparent face masks on speech signals.

Face masks muffle speech and make communication more difficult, especially for people with hearing loss. This study examines the acoustic attenuation caused by different face masks, including medical, cloth, and transparent masks, using a head-shaped loudspeaker and a live human talker. The results suggest that all masks attenuate frequencies above 1 kHz, that attenuation is greatest in front of the talker, and that there is substantial variation between mask types, especially cloth masks with different materials and weaves. Transparent masks have poor acoustic performance compared to both medical and cloth masks. Most masks have little effect on lapel microphones, suggesting that existing sound reinforcement and assistive listening systems may be effective for verbal communication with masks.

Acoustic effects of face masks on speech signals

Face masks muffle speech and make communication more difficult, especially for people with hearing loss. This study examines the acoustic attenuation caused by different face masks, including medical, cloth, and transparent masks, using a head-shaped loudspeaker and a live human talker. The results suggest that all masks attenuate frequencies above 1 kHz, that attenuation is greatest in front of the talker, and that there is substantial variation between mask types, especially cloth masks with different materials and weaves. Transparent masks have poor acoustic performance compared to both medical and cloth masks. Most masks have little effect on lapel microphones, suggesting that existing sound reinforcement and assistive listening systems may be effective for verbal communication with masks.

Identifying children's voices

Human adults rely on both acoustic and linguistic information to identify adult talkers. Assuming favorable conditions, adult listeners recognize other adults fairly accurately and quickly. But how well can adult listeners recognize child talkers, whose speech productions often differ dramatically from adult speech productions? Although adult talker recognition has been heavily studied, only one study to date has directly compared the recognition of unfamiliar adult and child talkers [Creel and Jimenez (2012). J. Exp. Child Psychol. 113(4), 487-509]. Therefore, the current study revisits this question with a much larger and younger sample of child talkers (N = 20); performance with adult talkers (N = 20) was also tested to provide a baseline. In Experiment 1, adults successfully distinguished between adult talkers in an AX discrimination task but performed much worse with child talkers. In Experiment 2, adults were slower and less accurate at learning to identify child talkers than adult talkers in a training-identification task. Finally, in Experiment 3, adults failed to improve at identifying child talkers after three days of training with numerous child voices. Taken together, these findings reveal a sizable difference in adults' ability to recognize child versus adult talkers. Possible explanations and implications for understanding human talker recognition are discussed.

Talker change detection: A comparison of human and machine performance.

The automatic analysis of conversational audio remains difficult, in part, due to the presence of multiple talkers speaking in turns, often with significant intonation variations and overlapping speech. The majority of prior work on psychoacoustic speech analysis and system design has focused on single-talker speech or multi-talker speech with overlapping talkers (for example, the cocktail party effect). There has been much less focus on how listeners detect a change in talker or in probing the acoustic features significant in characterizing a talker's voice in conversational speech. This study examines human talker change detection (TCD) in multi-party speech utterances using a behavioral paradigm in which listeners indicate the moment of perceived talker change. Human reaction times in this task can be well-estimated by a model of the acoustic feature distance among speech segments before and after a change in talker, with estimation improving for models incorporating longer durations of speech prior to a talker change. Further, human performance is superior to several online and offline state-of-the-art machine TCD systems.

Characterizing acoustic environments using spherical loudspeaker and microphone arrays

Room acoustic parameters, including the reverberation time (T60) and the direct-to-reverberant ratio (DRR), quantitatively describe the behaviour of an acoustic environment. These parameters are typically derived from an acoustic impulse response (AIR) obtained using an omnidirectional sound source and an omnidirectional receiver. However, the acoustic interaction between a source with a complex radiation pattern, for example, a human talker, and a receiver with a complex directivity pattern, for example, a human listener, cannot be accurately described by an omnidirectional AIR or the acoustic parameters derived from it. Here, we propose characterizing acoustic environments while taking the source and receiver directionality into account. A spherical loudspeaker array allows modelling a source with an arbitrary radiation pattern. Analogously, a spherical microphone can be used to describe a receiver with an arbitrary directivity pattern. By combining spherical loudspeaker and microphone arrays, a multiple-input multiple-output (MIMO) AIR of an acoustic environment can be measured to simulate sources and receivers with arbitrary directivities and estimate acoustic parameters accordingly.

A geometrical acoustic simulation of the effect of occupancy and source position in historical churches

Historical churches represent an important cultural heritage that nowadays is often used in different ways compared to the original design. In fact, both the “new style” of the celebration (introduced after the Second Vatican Council) and the frequent use as multifunctional spaces ask for different acoustic conditions, also changing the nature and the position of the sound sources compared with the original usage. Most of the available measurement campaigns were carried out using conventional sound positions and unoccupied conditions. However, due to the widespread use of scarcely sound absorbing surfaces, the presence of congregation may change dramatically the acoustical conditions inside the churches, and further variations can be introduced by changes in source placement. In order to analyze such effects, acoustic simulations were carried out using a commercial software, and starting from calibrated models of six different churches. The presence of the congregation determined a significant improvement in monaural acoustical parameters, especially in reverberation which was strongly reduced in all cases. Conversely, spatial parameters showed negligible variations. As variations in reverberation time were predicted with good accuracy by Sabine’s formula and variations in center time were in agreement with diffuse field theory, a simplified method to estimate occupied parameters from unoccupied values was proposed. Finally, the original acoustic conditions were simulated by placing a source with human talker directivity in front of the altar facing both the congregation and the main altar, and, where available, on the pulpit facing the congregation. When the priest turned his back to the congregation speech intelligibility dropped in all the churches, particularly where the reflecting surfaces of the apse were at greater distance. The pulpit position provided better results, however in most of the cases they were simply due to the shorter source–receiver distance.

Assessing the potential intelligibility of assistive listening systems.

Approximately 14% of the general population suffer from a noticeable degree of hearing loss and would benefit from some form of hearing assistance or deaf aid. Recent DDA legislation and requirements mean that many more assistive listening systems are being installed, yet there is evidence to suggest that many of these systems fail to perform adequately and provide the benefit expected. There has also been a proliferation of classroom and lecture room “soundfield” systems, with much conflicting evidence as to their apparent effectiveness. This paper reports on the results of some trial acoustic performance testing of such systems. In particular, the effects of system microphone type, distance, and location are shown to have a significant effect on the resultant performance. The potential of using the sound transmission index and, in particular, STIPa for carrying out installation surveys has been investigated, and a number of practical problems are highlighted. The requirements for a suitable acoustic test source to mimic a human talker are discussed. The findings discussed in the paper are also relevant to the installation and testing of “soundfield systems.”

Speech intelligibility improvement in rooms using microphone arrays.

Microphones arrays reduce the effects of reverberation and background noise as compared to a single microphone. The resulting improvement in speech intelligibility in rooms is investigated here, following the approach discussed by Houtgast et al. [Acustica 46, 60–72 (1980)]. Room characteristics (volume, reverberation time), the background noise spectrum, and the source-receiver separation are used to compute a modulation transfer function. This function is input into the procedure described in IEC standard 60268-16(2003) to arrive at an objective measure of speech intelligibility, the speech transmission index (STI). Doing this for both omnidirectional and microphone array receivers, the improvement in intelligibility is determined. The signal source has the spectral shape of a human talker, as given in the IEC standard; the background noise spectrum is based on measurements made in a restaurant. The benefit of using microphone arrays is discussed for various conditions.

Effect of Acoustic Environment on the Sensitivity of Speech Transmission Index to Source Directivity

Source directivity has the potential to affect speech transmission index (STI) measurements. When the source is a model of a human talker, there is a question of how accurate that model needs to be. However, in many acoustic environments source directivity has little effect on STI because other factors dominate. One instance is soundfields in which the direct sound is dominant (e.g., in the quiet outdoors, in the nearfield, or in an anechoic room): directivity will have no effect on STI so long as the direct sound is strong enough. Another instance is soundfields in which the reverberant soundfield is dominant (e.g., in the far field in a room with moderate or more reverberation). This paper examines theoretical situations where source directivity has a substantial influence on STI because of the balance between direct (or early) and reverberant soundfields, as well as the role that background noise can have in increasing the importance of source directivity.

The role of source and filter characteristics in human talker identification: Experiments with laryngeal and electrolarynx speech.

Differences in individuals’ vocal anatomy and physiology result in unique acoustic features of their vocalizations. Humans are exceptionally attuned to these variations and use them to identify familiar individuals. Although these abilities are often called “voice recognition,” talker identity cues actually arise through interactions between acoustic excitation produced at the source (typically, the larynx) and both static and dynamic properties of the filter (vocal tract, articulators, and their manipulations during speech). We investigated the differential contributions of source‐ and filter‐related information to talker identification through four experiments using laryngeal (typical) and electrolarynx speech from five talkers. Using an electrolarynx energy source removed individual differences in vocal anatomy, leaving only unique filter properties for talker identification. Listeners learned talker identity best from typical, laryngeal speech, which contained both unique source and filter cues. Liste...

Influences of perceived racial identity on human talker identification

Talker identification, the process by which human listeners recognize individuals by their voice, is one of the most poorly understood abilities of the human auditory system. Current psychological models of talker identification rely on strict analogies to face perception and the visual system, despite differences in the objects of perception in these two modalities. Here we investigate the existence of an own‐race bias in voice perception ‐ a phenomenon which has been pivotal in the study of face perception. Our results demonstrate an own‐race bias in talker identification: listeners of different ethnic backgrounds show an advantage for identifying individual voices of the same race as themselves. However, unlike in vision, the own‐race bias in talker identification manifests based specifically on the perceived, but not actual, race of a talker. The influence of perceived race suggests physical (voice structural) cues do not give rise to this effect. Instead, the own‐race bias in talker identification is a result of listeners' asymmetric exposure to talkers' socially‐acquired manners of expression (i.e. the dynamic features of voice and speech). Such manners of expression may be stereotypically associated with a particular ethnic group, although not actually exhibited by all members of that group. [Work supported by NIH]

Visual Phonemic Ambiguity and Speechreading

To study the role of visual perception of phonemes in visual perception of sentences and words among normal-hearing individuals. Twenty-four normal-hearing adults identified consonants, words, and sentences, spoken by either a human or a synthetic talker. The synthetic talker was programmed with identical parameters within phoneme groups, hypothetically resulting in simplified articulation. Proportions of correctly identified phonemes per participant, condition, and task, as well as sensitivity to single consonants and clusters of consonants, were measured. Groups of mutually exclusive consonants were used for sensitivity analyses and hierarchical cluster analyses. Consonant identification performance did not differ as a function of talker, nor did average sensitivity to single consonants. The bilabial and labiodental clusters were most readily identified and cohesive for both talkers. Word and sentence identification was better for the human talker than the synthetic talker. The participants were more sensitive to the clusters of the least visible consonants with the human talker than with the synthetic talker. It is suggested that ability to distinguish between clusters of the least visually distinct phonemes is important in speechreading. Specifically, it reduces the number of candidates, and thereby facilitates lexical identification.

Investigating the audio–visual speech detection advantage

Seeing the moving face of the talker permits better detection of speech in noise compared to auditory only presentation, an Audio–Visual (AV) facilitation effect. Three experiments that used a masked speech detection task are reported. The experiments were designed to contrast two accounts of the AV facilitation effect (AV peak listening and AV grouping). In each experiment a different manipulation of the relationship between the auditory and visual signals was employed. The first experiment manipulated the sequencing of the visual and auditory information by presenting the displays time normal or time reversed. The results showed that AV facilitation only occurred for the time normal presentations where there was a high correlation between the AV signals. Experiment 2 examined the impact on AV facilitation of shifting the auditory signals earlier in time than its normal position (again with time normal and time reversed presentation). It was found that shifting the auditory component abolished the AV effect for both the time normal and reversed conditions. The final experiment examined the AV detection advantage using another situation in which the relationship between the AV signals differed. Two versions of AV speech produced by a virtual talker were investigated. In one version, based on text-to-speech synthesis, the video and auditory signals were more rapid than the human talker of Experiment 1. In the other version, the signals were lengthened to match the durations of the human talker. A small but reliable AV facilitation effect was only found for the second version. The results are consistent with a cross-modal peak listening account and are discussed in terms of constraints on the integration of auditory and visual speech.

Correlation and Stationarity of Speech Radiation: Consequences for Linear Multichannel Filtering

Speech processing using multichannel microphone systems is often based on slowly adapting, linear filters. These systems are able to extract a single source from a mixture (and suppress the others)-if the speech radiation can be described by a linear and time-invariant transfer function. Here, we test this assumption using a two-channel microphone array and a human talker as the speech source. We measure correlations between the signals received by the two microphones for individual phonemes using the magnitude squared coherence. Stationarity is addressed by comparing optimal filters between different phoneme pairs using the system distance. We find that, in particular for fricatives, the coherence of the speech signals radiated to different directions is very low. We also find, that the transfer functions from the mouth to the microphones differ significantly between vowels, depending on the locations of the two microphones. These measurements show that the general mixing model does not hold for speech for arbitrary microphone setups, and that multichannel microphone systems have to be carefully designed.

Directivity of human talkers

As well as the level of speech, the directivity of the sound field surrounding the human head is important when estimating speech privacy in open offices. Information on this subject was found to be sparse so the sound field surrounding human talkers was recorded by two sets of eight microphones fixed on two orthogonal meridian arcs in an anechoic chamber. Talkers were positioned at the center of the two arcs of the microphone arrays and rotated through six positions at 15° increments to survey the whole sound field. Twenty male and twenty female subjects were used. No significant differences in directivity were detected between male and female talkers although the mean spectra were different. There was good agreement between the field and laboratory measurements of the male voice spectra though some differences were found for the female voice spectra. No significant differences in directivity or frequency content were detected between English and French talkers. Similar directivities were obtained for the average human talker and the Bruel and Kjaer Head and Torso Simulator. Agreement with other published information is fairly good. This work was done as part of a project for Public Works and Government Services Canada.