Acoustic Input Research Articles

Cortical speech tracking is related to individual prediction tendencies.

Listening can be conceptualized as a process of active inference, in which the brain forms internal models to integrate auditory information in a complex interaction of bottom-up and top-down processes. We propose that individuals vary in their "prediction tendency" and that this variation contributes to experiential differences in everyday listening situations and shapes the cortical processing of acoustic input such as speech. Here, we presented tone sequences of varying entropy level, to independently quantify auditory prediction tendency (as the tendency to anticipate low-level acoustic features) for each individual. This measure was then used to predict cortical speech tracking in a multi speaker listening task, where participants listened to audiobooks narrated by a target speaker in isolation or interfered by 1 or 2 distractors. Furthermore, semantic violations were introduced into the story, to also examine effects of word surprisal during speech processing. Our results show that cortical speech tracking is related to prediction tendency. In addition, we find interactions between prediction tendency and background noise as well as word surprisal in disparate brain regions. Our findings suggest that individual prediction tendencies are generalizable across different listening situations and may serve as a valuable element to explain interindividual differences in natural listening situations.

Comparison of Acoustic Noise and Vibration in Ball-Bearing-Supported Motors and One-Axis Actively Positioned Single-Drive Bearingless Motor With Two Radial Permanent-Magnet Passive Magnetic Bearings

This study experimentally investigates the acoustic noise, vibration, and power consumption in a one-degree-of-freedom actively positioned single-drive bearingless motor, which has radial passive magnetic bearings (RPMBs) and compared to an identical stator part and rotor shaft with radial mechanical ball bearings. For the experiment, three test motors were set up: (a) a bearingless motor with two RPMB, (b) a motor with two ball bearings without an axial preload, and (c) a motor with two ball bearings with an axial preload. Motor (a) under test had one-axis active positioning and the radial movements were supported by RPMB made of cylindrical permanent magnets. Conversely, in motors (b) and (c), the radial and axial movements were supported by ball bearings, and there was no production of active axial force. The experimental results confirmed that the levels of acoustic noise, stator vibration, and input power consumption were significantly lower in motor (a) than those in motors (b) and (c). In the analysis section, dynamic models of the bearingless motor with RPMB and motor with ball bearings were designed and simulated using MATLAB <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\backslash$</tex-math></inline-formula> Simulink. The low radial stiffness in RPMB was found to contribute to acoustic noise and vibration reductions. Thus, this article presents an example of a one-degree-of-freedom actively positioned bearingless motor with RPMB that realizes reductions of acoustic noise, stator vibration, and input power consumption.

Parametric model of young infants’ eardrum and ear canal impedances supporting immittance measurement results. Part II: Prediction of eardrum and ear canal impedances for frequent pathological middle ear conditions

In order to gain a better understanding of wideband acoustic immitance (WAI) measurements, in this second part of a two-part paper, the parametric electro-acoustic model of the ear canal and the middle ear of young infants proposed in the first part is extended. The extension allows predictions of the influence of the pathological middle ear conditions middle ear effusion and negative static air pressure difference between the middle ear and the ear canal. Comparisons of the acoustic input impedance of the ear predicted by the model with real ear measurements in young infants’ ears with middle ear effusion show that the effects due to the pathology can be predicted well. For the negative static air pressure, a modeling approach was proposed but could not be confirmed yet, due to a lack of available measurement data. Furthermore, comparisons between different middle ear states (healthy, middle ear effusion and static air pressure difference) predicted by the model showed characteristic differences in all relevant WAI measures. However, it is also shown that WAI measures requiring an estimate of the cross-sectional area at the measurement position, i.e., absorbance and reflectance, are highly sensitive to this estimate.

Audiological and Demographic Factors that Impact the Precision of Speech Categorization in Cochlear Implant Users.

The ability to adapt to subtle variations in acoustic input is a necessary skill for successful speech perception. Cochlear implant (CI) users tend to show speech perception benefits from the maintenance of their residual acoustic hearing. However, previous studies often compare CI users in different listening conditions within-subjects (i.e., in their typical Acoustic + Electric configuration compared with Acoustic-only or Electric-only configurations) and comparisons among different groups of CI users do not always reflect an Acoustic + Electric benefit. Existing work suggests that CI users with residual acoustic hearing perform similarly to Electric-only listeners on phonetic voicing contrasts and unexpectedly poorer with fricative contrasts which have little energy in the range of the Acoustic + Electric listeners' acoustic hearing. To further investigate how residual acoustic hearing impacts sensitivity to phonetic ambiguity, we examined whether device configuration, age, and device experience influenced phonetic categorization in a large individual differences study. CI users with various device configurations (Electric-only N = 41; Acoustic + Electric N = 95) categorized tokens from five /b-p/ and five /s-ʃ/ minimal pair continua (e.g., bet-pet; sock-shock). We investigated age, device experience, and when applicable, residual acoustic hearing (pure tone hearing thresholds) as predictors of categorization. We also examined the relationship between phonetic categorization and clinical outcomes (CNC, AzBio) in a subset of our sample. Acoustic + Electric CI users were better able to categorize along the voicing contrast (steeper categorization slope) compared with Electric-only users, but there was no group-level difference for fricatives. There were differences within the subgroups for fricatives: bilateral users showed better categorization than unilateral users and bimodal users had better categorization than hybrid users. Age was a significant factor for voicing, while device experience was significant for fricatives. Critically, within the Acoustic + Electric group, hybrid CI users had shallower slopes than bimodal CI users. Our findings suggest residual acoustic hearing is beneficial for categorizing stop voicing, but not frication. Age impacts the categorization of voicing, while device experience matters for fricatives. For CI users with ipsilateral residual acoustic hearing, those with better hearing thresholds may be over-relying on their acoustic hearing rather than extracting as much information as possible from their CI, and thus have shallower fricative categorization.

Elastic Piezoelectric Nanofibers Mats for Acoustic Energy Harvesting

One of the traditional clean-energy harvesting solutions is through transducing different mechanical stresses into electrical energy. Generally, the acoustic-to-electric energy conversion of still needs more research investigations to be applicable. In our work, we are targeting to fabricate elastic nanofibers mats via electrospinning method to be used for acoustic harvesting/sensing applications. The targeted mechanically-elastic nanocomposite includes polyvinylidene fluoride (PVDF), which is one of the most famous organic piezo materials, with blended thermoplastic polyurethane (TPU). As TPU supports higher mechanical allowed breaking strain. Then, the synthesized mat has been used as a target for mechanical stresses with resulted piezosensitivity of 667±220 mV/N. Then, the nanofibers mat has been targeted against acoustic signals with different amplitude and frequencies. It has been observed that the synthesized mats can detect or harvest acoustic signals and convert them into output electric voltage. According to acoustic sound input, the synthesized electrospun nanofibers detect output voltage up to 300 mV with increased input audible amplitude and frequency up to 6 kHz, where the harvested voltage has a saturation behaviour beyond that audible frequency. That can open the track for using such nanocomposites in energy harvesting applications from disposable facemasks, filters, and music/noise in different opened and closed areas.

Experimental study of acoustic agglomeration coupled with water droplets on eliminating cable fire smoke

Acoustic agglomeration to eliminate fire smoke is one of the most promising technologies at this stage, and we effectively solve technical stumbling blocks such as narrow acoustic frequency range, sound pressure level injury, and long agglomeration time using acoustic waves coupled with water droplets. According to the mechanism and experimental research, acoustic waves coupled with water droplets technology can broaden the optimal agglomeration frequency range, reduce the influence of acoustic power, and effectively improve agglomeration efficiency. Water droplets with a concentration of 40 g/m 3 were added at the acoustic frequency of 1.5 kHz and the acoustic power of 10 W to achieve the safe escape threshold for personnel within 30 s of working while reducing the damage to the human ear caused by the high sound pressure level. • A new method to eliminate fire smoke is proposed. • The addition of water droplets can expand the range of agglomeration frequencies. • 50% reduction in required agglomeration time. • The addition of water droplets can reduce the acoustic power input. • Reduces damage to the human ear from high sound pressure levels.

Entropy transfer function measurement with tunable diode laser absorption spectroscopy

Tunable diode laser absorption spectroscopy (TDLAS) with wavelength modulation spectroscopy (WMS) has been widely applied to measure the gas temperature and species concentration in combustion environments. In this work, we show the capability of TDLAS in measuring the coherent temperature fluctuations of a turbulent swirled flame. Two distributed feedback grating (DFB) lasers are used to probe the H2O absorption transitions near 7185.59 cm−1 and 6806.03 cm−1. Using WMS, the 2f/1f absorption spectra are converted into temperature at a rate of 5 kHz. The quantity of interest that we measure is the Entropy Transfer Function (ETF) which, for low Mach numbers, relates the upstream acoustic perturbation to the temperature fluctuation downstream of the flame. The ETF is obtained by taking the ratio between the measured temperature fluctuation downstream of the flame, measured with TDLAS and the upstream acoustic perturbation input, induced by loudspeakers with forcing frequency ranging from 40 to 250 Hz, reconstructed with the multi-microphone method. We show that the TDLAS-WMS technique can measure relatively low coherent temperature fluctuations as low as 5 K. It is found that the gain of ETFs of the swirling flame does not decrease monotonically with respect to the forcing frequency and exhibits a minimum response at particular frequencies depending on the operating conditions. Moreover, the ETFs for different operating conditions collapse on each other when plotted against the Strouhal number. The similarity between the measured ETFs and Flame Transfer Functions (FTFs) indicates a strong link between flame heat release rate response and the temperature fluctuation downstream of the flame. Furthermore, the measurement series demonstrate the strong capability and suitability of the TDLAS-WMS technique for entropy waves measurements, leading to the discovery of peculiar entropy responses at high frequencies.

Effects of an enhanced acoustic environment on residual hearing following chronic cochlear implantation and electrical stimulation in the partially deafened cat

There is an increasing trend to provide cochlear implants for people with useful residual hearing, typically in the low frequency range (<2 kHz). These recipients typically use both electrical stimulation from their implant and acoustic stimulation that has been amplified with a hearing aid to access their residual hearing, so called electro-acoustic stimulation (EAS). However, a significant problem is the loss of residual hearing following implantation that can occur immediately following surgery or delayed over many months. One potential cause of the loss of residual hearing is the over stimulation of remaining hair cells due to the combination of an amplified acoustic input and direct electrical activation. This paper aims to test this hypothesis.Here, we have used a neonatal aminoglycoside-induced partial hearing cat model that resulted in a high frequency hearing loss (>4 kHz). Two separate cohorts of animals were implanted and received unilateral chronic electrical stimulation using clinical stimulators and speech processors over 5 months. To simulate potential over stimulation via a hearing aid, one cohort of animals were also exposed to an enhanced acoustic environment consisting of 80 dB SPL 4-talker babble presented 14 h per day. Hearing thresholds for both stimulated and unstimulated ears were measured throughout the implantation period. Cochleae were collected for histology to measure spiral ganglion neuron survival, hair cell survival and tissue response to chronic implantation and electrical stimulation. Consistent with clinical observations, cochlear implantation and stimulation resulted in an increase in threshold across the population. There was no significant effect of the enhanced acoustic environment on auditory thresholds or histological measures (hair cell survival, neuronal survival) of hearing, indicating that hair cell overstimulation was not a significant driver of loss of residual hearing.

Modeling fast acoustic streaming: Steady-state and transient flow solutions.

Traditionally, acoustic streaming is assumed to be a steady-state, relatively slow fluid response to passing acoustic waves. This assumption, the so-called slow streaming assumption, was made over a century ago by Lord Rayleigh. It produces a tractable asymptotic perturbation analysis from the nonlinear governing equations, separating the acoustic field from the acoustic streaming that it generates. Unfortunately, this assumption is often invalid in the modern microacoustofluidics context, where the fluid flow and acoustic particle velocities are comparable. Despite this issue, the assumption is still widely used today, as there is no suitable alternative. We describe a mathematical method to supplant the classic approach and properly treat the spatiotemporal scale disparities present between the acoustics and remaining fluid dynamics. The method is applied in this work to well-known problems of semi-infinite extent defined by the Navier-Stokes equations, and preserves unsteady fluid behavior driven by the acoustic wave. The separation of the governing equationsbetween the fast (acoustic) and slow (hydrodynamic) spatiotemporal scales are shown to naturally arise from the intrinsic properties of the fluid under forcing, not by arbitrary assumption beforehand. Solution of the unsteady streaming field equationsprovides physical insight into observed temporal evolution of bulk streaming flows that, to date, have not been modeled. A Burgers equationis derived from our method to represent unsteady flow. By then assuming steady flow, a Riccati equationis found to represent it. Solving these equationsproduces direct, concise insight into the nonlinearity of the acoustic streaming phenomenon alongside an absolute, universal upper bound of 50% for the energy efficiency in transducing acoustic energy input to the acoustic streaming energy output. Rigorous validation with respect to experimental and theoretical results from the classic literature is presented to connect this work to past efforts by many authors.

Acoustic-based automatic speech intelligibility scoring using deep neural networks

Human-generated measures of speech intelligibility are time-intensive methods for assessing the intelligibility of speech. The purpose of the present study is to automate the assessment of speech intelligibility by developing a deep neural network that estimates a standardized intelligibility score based on acoustic input. We extracted Mel-frequency cepstral coefficients from the UW/NU IEEE sentence corpus which had been manipulated with three signal-to-noise ratios (−2, 0, 2 dB). We obtained listener transcriptions from the UAW speech intelligibility dataset and calculated the Levenshtein distance between the transcriptions and the speaker's prompt. The neural network was trained to predict the Levenshtein distance given MFCC representations of sentences. We use tenfold cross-validation to verify the accuracy of the model and investigate the correlation of the model predictions with the average human responses. We also compare our model’s accuracy with the Levenshtein distance generated by transcriptions produced by the DeepSpeech ASR model. This study investigates the reliability of deep neural networks as an alternative to human-based inference in quantifying the intelligibility of speech. We discuss the advantages and disadvantages of the different approaches to assessing speech intelligibility.

Spectral changes to sound exposure when preterm infants transition from the incubator to an open crib

Children born preterm are at risk for speech and language developmental delays and disorders. It has been proposed that adverse auditory exposures during stay in the neonatal intensive care unit (NICU) may contribute to this risk, as it is well-established that the preterm infant’s developing auditory system is sensitive to acoustic input during this time. While reduced speech exposure, noxious noise levels, and excessive silence in the NICU are of concern, another potential cause for concern is abrupt changes in auditory exposure during NICU stay. Our previous data indicate that NICU incubator walls have a low-pass filtering effect, attenuating external sounds at frequencies above 200 Hz. However, internal sounds generated by the incubator or other life-saving devices still produce noise exposure. Furthermore, when infants transition to an open crib from the incubator based on improved overall health, they may be at risk of increased exposure to higher frequencies. Here we present spectral analyses of auditory exposure recordings made for several preterm infants throughout NICU stay. Our analysis reveals that for frequencies above 500 Hz, sound levels are significantly higher in the open crib; below 500 Hz, levels are generally higher in the incubator. These data point to yet another potentially disruptive effect of the NICU environment caused by an abrupt change in auditory exposure.

Maintenance of phonological contrast in speech production of cochlear implant users

Prior work has shown that normal-hearing (NH) talkers produce more distinct vowels in words having a lower word frequency (WF) and higher neighborhood density (ND). Phonological contrast in speech production may differ between post-lingually deafened cochlear implant (CI) users and individuals with NH due to the degraded auditory input provided by CIs. The current study sought to determine whether the same pattern of vowel production exists in CI users. CI users were recorded reading monosyllabic words that were lexically “hard” (low WF, high ND) or “easy” (high WF, low ND). For each participant, vowel dispersion was calculated as the Euclidian distance from the center of the acoustic vowel space. If degraded acoustic input affects phonological contrast in CI users, we would not expect vowel dispersion to differ between vowels in “hard” and “easy” words, contrary to talkers with NH. Preliminary results revealed a significant effect of lexical difficulty on vowel dispersion, indicating that CI users produced more distinct vowels in “hard” words than in “easy” words. These preliminary findings could suggest that degraded auditory input does not substantially affect phonological contrast in vowel production in CI users, or that long-term adaptation to CI input can restore contrast after auditory deprivation.

Phonetic category activation predicts the direction and magnitude of perceptual adaptation to accented speech.

Unfamiliar accents can systematically shift speech acoustics away from community norms and reduce comprehension. Yet, limited exposure improves comprehension. This perceptual adaptation indicates that the mapping from acoustics to speech representations is dynamic, rather than fixed. But, what drives adjustments is debated. Supervised learning accounts posit that activation of an internal speech representation via disambiguating information generates predictions about patterns of speech input typically associated with the representation. When actual input mismatches predictions, the mapping is adjusted. We tested two hypotheses of this account across consonants and vowels as listeners categorized speech conveying an English-like acoustic regularity or an artificial accent. Across conditions, signal manipulations impacted which of two acoustic dimensions best conveyed category identity, and predicted which dimension would exhibit the effects of perceptual adaptation. Moreover, the strength of phonetic category activation, as estimated by categorization responses reliant on the dominant acoustic dimension, predicted the magnitude of adaptation observed across listeners. The results align with predictions of supervised learning accounts, suggesting that perceptual adaptation arises from speech category activation, corresponding predictions about the patterns of acoustic input that align with the category, and adjustments in subsequent speech perception when input mismatches these expectations. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

The Effect of Feed Frequency on Growth, Survival and Behaviour of Juvenile Spiny Lobster (Panulirus ornatus).

Simple SummarySensory stimuli including olfactory, visual, acoustic and tactile input, are important aspects of feed exposure and optimisation of feeding schedules. These aspects were investigated in our study, with a focus on the effect of feed frequency (one, two, four, eight, sixteen and thirty-two feed supplies per day) on survival, growth and behaviour of juvenile P. ornatus. Increasing daily feed frequency from one to sixteen improved feed intake and growth. Spiny lobsters have a range of complex chemical communication pathways that contribute to feeding behaviour. Feed intake is modulated by feed availability and feed characteristics, such as attractiveness and palatability, with behavioural factors, such as social competition and circadian rhythm, providing an extra layer of complexity. In this study, we investigated the effect of feed frequency on survival and growth of early-stage (instar 2–6) juvenile Palunirus ornatus. In addition, we investigated the interactive effect of feed frequency and circadian rhythm on lobster feed response. Lobsters were fed a set ration at a frequency of either one, two, four, eight, sixteen or thirty-two times per day over 49 days. The effect of feed frequency on growth and survival was determined. Circadian feeding activity under these feeding treatments was assessed by time-lapse photography. Increased feed frequency from one to sixteen feeds daily improved growth by increasing apparent feed intake (AFI) and feed attraction, as confirmed by the increased presence of lobsters in the feeding area. The rapid leaching of feed attractant, particularly free amino acid, suggests a beneficial effect of multiple feeding frequencies on feed intake and growth. However, more than sixteen feeds per day resulted in decreased feed intake and a subsequent reduction in growth. The decrease in feed intake is thought to be associated with saturation of the culture environment with attractants, resulting in a reduced behavioural response to feed supplies. This may indicate the need for depletion of attractants to retrigger a feeding response. As lobsters were grown communally, faster growth at sixteen rations per day was also coupled with increased cannibalism, likely driven by increased vulnerability with the occurrence of more frequent ecdysis events. Whereas circadian rhythm indicated more activity at night, an interaction between daytime activity and feed frequency was not observed.

An ERP investigation of accented isolated single word processing

Previous studies show that there are differences in native and non-native speech processing (Lev-Ari, 2018). However, less is known about the differences between processing native and dialectal accents. Is dialectal processing more similar to foreign or native speech? To address this, two theories have been proposed. The Perceptual Distance Hypothesis states that the mechanisms underlying dialectal accent processing are attenuated versions of those of foreign (Clarke & Garrett, 2004). Conversely, the Different Processes Hypothesis argues that the mechanisms of foreign and dialectal accent processing are qualitatively different (Floccia et al., 2009). The present study addresses these hypotheses. Electroencephalographic data was recorded from 25 participants who listened to 40 isolated words in different accents. Event-Related Potential mean amplitudes were extracted: P2 [150–250 ms], PMN [250–400 ms] and N400 [400–600 ms]. Support for the Different Processes Hypothesis was found in different time windows. Results show that early processing mechanisms distinguish only between native and non-native speech, with a reduced P2 amplitude for foreign accent processing, supporting the Different Processes Hypothesis. Furthermore, later processing mechanisms show a similar binary difference in the processing of the accents, with a larger PMN negativity elicited in the foreign accent than the others, further supporting the Different Processes Hypothesis. Results contribute to the understanding of single word processing, in which it is uniquely difficult to extract acoustic characteristics from foreign accent, and in which foreign accented speech is associated with the largest cost, as compared to native and dialectal speech, of phonological matching between representations and acoustic input.

Silent Speech Interface: An Inversion Problem

When conventional acoustic-verbal communication is neither possible or desirable, silent speech interfaces (SSI) rely on biosignals, non-acoustic signals created by the human body during speech production, to facilitate communication. Despite considerable advances in sensing techniques that can be employed to capture these biosignals, majority of them are used under controlled scenarios in laboratories. One such example is Electromagnetic Articulograph (EMA), which monitors articulatory motion. It is expensive with inconvenient wiring and practically not portable in real world. Since articulator measurement is difficult, articulatory parameters may be estimated from acoustics through inversion. Acoustic-to-articulatory inversion (AAI) is a technique for determining articulatory parameters using acoustic input. Automatic voice recognition, text-to- speech synthesis, and speech accent conversion can all benefit from this. However, for speakers with no articulatory data, inversion is required in many practical applications. Articulatory reconstruction is more useful when the inversion is speaker independent. Initially, we analysed positional data to better understand the relationship between sensor data and uttered speech. Following the analysis, we built a speaker independent articulatory reconstruction system that uses a Bi- LSTM model. Additionally, we evaluated the trained model using standard evaluation measures.

Short-term perceptual reweighting in suprasegmental categorization

Segmental speech units such as phonemes are described as multidimensional categories whose perception involves contributions from multiple acoustic input dimensions, and the relative perceptual weights of these dimensions respond dynamically to context. For example, when speech is altered to create an “accent” in which two acoustic dimensions are correlated in a manner opposite that of long-term experience, the dimension that carries less perceptual weight is down-weighted to contribute less in category decisions. It remains unclear, however, whether this short-term reweighting extends to perception of suprasegmental features that span multiple phonemes, syllables, or words, in part because it has remained debatable whether suprasegmental features are perceived categorically. Here, we investigated the relative contribution of two acoustic dimensions to word emphasis. Participants categorized instances of a two-word phrase pronounced with typical covariation of fundamental frequency (F0) and duration, and in the context of an artificial “accent” in which F0 and duration (established in prior research on English speech as “primary” and “secondary” dimensions, respectively) covaried atypically. When categorizing “accented” speech, listeners rapidly down-weighted the secondary dimension (duration). This result indicates that listeners continually track short-term regularities across speech input and dynamically adjust the weight of acoustic evidence for suprasegmental decisions. Thus, dimension-based statistical learning appears to be a widespread phenomenon in speech perception extending to both segmental and suprasegmental categorization.

Control of Acoustic Energy Input for Cleaning of Industrial Boilers

A non-intrusive cleaning method for boiler tubes at Sasol Synfuels power station at Secunda, in the Mpumalanga province of South Africa, is preferred over conventional methods that require boiler shutdown. The elected non-intrusive cleaning method utilizes sound energy waves, produced by an acoustic horn. Due to the nature of sound propagation and the effectiveness required, there is a requisite to control and operate the sonic horn. If the acoustic horn’s sound frequency is too low, it will produce higher sound energy waves that will resonate with the plant’s harmonious frequency and cause structural damage. Conversely, if the sonic horn’s sound frequency is too high, excessive noise levels may be reached and annoy plant personnel. To prevent these undesirable outcomes posed by adopting acoustic cleaning, there needs to be a regulatory system incorporated into the configuration to mitigate vibrations and limit noise. The regulatory system comprises a control system that drives the acoustic horn’s sound frequency as intended through a set point. The designed control system meets the anticipated requirements, such that it has an ideal transient response of 0.562 s, a steady-state error achieved in 1.05 s, with 0.201% overshoot, and most importantly the closed-loop system is stable.

Efficient acoustic feature transformation in mismatched environments using a Guided-GAN

We propose a new framework to improve automatic speech recognition (ASR) systems in resource-scarce environments using a generative adversarial network (GAN) operating on acoustic input features. The GAN is used to enhance the features of mismatched data prior to decoding, or can optionally be used to fine-tune the acoustic model. We achieve improvements that are comparable to multi-style training (MTR), but at a lower computational cost. With less than one hour of data, an ASR system trained on good quality data, and evaluated on mismatched audio is improved by between 11.5% and 19.7% relative word error rate (WER). Experiments demonstrate that the framework can be very useful in under-resourced environments where training data and computational resources are limited. The GAN does not require parallel training data, because it utilises a baseline acoustic model to provide an additional loss term that guides the generator to create acoustic features that are better classified by the baseline.

Effects of tonotopic matching and spatial cues on segregation of competing speech in simulations of bilateral cochlear implants.

In the clinical fitting of cochlear implants (CIs), the lowest input acoustic frequency is typically much lower than the characteristic frequency associated with the most apical electrode position, due to the limited electrode insertion depth. For bilateral CI users, electrode positions may differ across ears. However, the same acoustic-to-electrode frequency allocation table (FAT) is typically assigned to both ears. As such, bilateral CI users may experience both intra-aural frequency mismatch within each ear and inter-aural mismatch across ears. This inter-aural mismatch may limit the ability of bilateral CI users to take advantage of spatial cues when attempting to segregate competing speech. Adjusting the FAT to tonotopically match the electrode position in each ear (i.e., increasing the low acoustic input frequency) is theorized to reduce this inter-aural mismatch. Unfortunately, this approach may also introduce the loss of acoustic information below the modified input acoustic frequency. The present study explored the trade-off between reduced inter-aural frequency mismatch and low-frequency information loss for segregation of competing speech. Normal-hearing participants were tested while listening to acoustic simulations of bilateral CIs. Speech reception thresholds (SRTs) were measured for target sentences produced by a male talker in the presence of two different male talkers. Masker speech was either co-located with or spatially separated from the target speech. The bilateral CI simulations were produced by 16-channel sinewave vocoders; the simulated insertion depth was fixed in one ear and varied in the other ear, resulting in an inter-aural mismatch of 0, 2, or 6 mm in terms of cochlear place. Two FAT conditions were compared: 1) clinical (200–8000 Hz in both ears), or 2) matched to the simulated insertion depth in each ear. Results showed that SRTs were significantly lower with the matched than with the clinical FAT, regardless of the insertion depth or spatial configuration of the masker speech. The largest improvement in SRTs with the matched FAT was observed when the inter-aural mismatch was largest (6 mm). These results suggest that minimizing inter-aural mismatch with tonotopically matched FATs may benefit bilateral CI users’ ability to segregate competing speech despite substantial low-frequency information loss in ears with shallow insertion depths.