Nonlinear auditory models yield new insights into representations of vowels

Carney, Laurel H.; McDonough, Joyce

doi:10.3758/s13414-018-01644-w

Cited by 18 publications

(14 citation statements)

References 57 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This is indeed consistent with demonstrations of nonlinear processing in the auditory periphery. 57 Given that both DNN and feature encoding models have roughly similar amounts of predictors (on the order of 100), this suggests that DNNs are learning nonlinear representations that are critical for extracting relevant phonetic features. 2) The dynamic temporal integration of contextual information in the DNN models: this is particularly critical for higher-order speech responses in the non-primary auditory cortex.…”

Section: Discussionmentioning

confidence: 99%

Dissecting neural computations of the human auditory pathway using deep neural networks for speech

Anumanchipalli

Mohamed

et al. 2022

Preprint

View full text Add to dashboard Cite

The human auditory system extracts rich linguistic abstractions from the speech signal. Traditional approaches to understand this complex process have used classical linear feature encoding models, with limited success. Artificial neural networks have recently achieved remarkable speech recognition performance and offer potential alternative computational models of speech processing. We used the speech representations learned by state-of-the-art deep neural network (DNN) models to investigate neural coding across the ascending auditory pathway from the peripheral auditory nerve to auditory speech cortex. We found that representations in hierarchical layers of the DNN correlated well to neural activity throughout the ascending auditory system. Unsupervised speech models achieve the optimal neural correlations among all models evaluated. Deeper DNN layers with context-dependent computations were essential for populations of high order auditory cortex encoding, and the computations were aligned to phonemic and syllabic context structures in speech. Accordingly, DNN models trained on a specific language (English or Mandarin) predicted cortical responses in native speakers of each language. These results reveal convergence between representations learned in DNN models and the biological auditory pathway and provide new approaches to modeling neural coding in the auditory cortex.

show abstract

Section: Discussionmentioning

confidence: 99%

Dissecting neural computations of the human auditory pathway using deep neural networks for speech

Anumanchipalli

Mohamed

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…The inattention to auditory representation is linked to the lack of realistic models of the neural responses to speech in the auditory pathways. Realistic neural models have been developed to model the coding of speech in the midbrain (Carney, 2018; Carney and McDonough, 2019; Nelson and Carney, 2004; Zilany et al, 2014; Zilany et al, 2009).…”

Section: Discussionmentioning

confidence: 99%

“…The AN model provided the input for both types of IC model in the form of time-varying rate, convolved with functions representing excitatory or inhibitory postsynaptic responses, which differed for the two cell types (Carney et al, 2015). The postsynaptic potential time constants and delays were set to produce BE responses with a BMF of 100 Hz (Carney and McDonough, 2019). This BMF is near the center of the distribution of IC BMFs (Kim et al, 2020; Krishna and Semple, 2000; Nelson and Carney, 2007).…”

Section: Methodsmentioning

confidence: 99%

Representations of fricatives in sub-cortical model responses: comparisons with human consonant perception

Hamza

Farhadi

Schwarz³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Fricatives are obstruent phonemes made by constricting airflow in the vocal tract, producing turbulence across the constriction site or downstream from the constriction. Fricatives exhibit significant intra/inter-subject and contextual variability. Yet fricatives are perceived with high accuracy. The current study investigated modeled neural responses to fricatives in the auditory nerve (AN) and inferior colliculus (IC), with the hypothesis that response profiles across populations of neurons provide robust correlates to perception. Stimuli were 270 intervocalic fricatives (10 speakers x 9 fricatives x 3 utterances). Computational model response profiles had 50 characteristic frequencies, log-spaced from 125 to 8 or 20 kHz. Confusion matrices generated by KNN subspace classifiers used the average rates for the 50 characteristic frequencies as the features and compared them with published behavioral data. The modeled AN and IC neural responses provided better predictions of perceptual accuracy than the stimulus spectra, with IC showing better accuracy than AN. Perceptual fricative accuracy was explained by modeled neural response profiles, whereas confusions were only partially explained. Extended frequencies improved accuracy based on the model IC, corroborating the importance of extended high frequencies in speech perception. Future electrophysiological studies from IC cells and neural response simulations of hearing-impaired ears could guide the development of processing strategies.

show abstract

“…A direct comparison between the linearly ramping shape used in this study and the exponentially ramped shape used in Yip et al is therefore not straightforward due to different stimulation parameters tested. However, one explanation for the larger effect observed by Yip et al could be that an exponentially ramped shape is simply more charge-efficient than the linearly ramping shapes used with rampUp and rampDown, possibly because the auditory system, and biology in general, operate on non-linear scales (Carney & McDonough, 2019). Another possible explanation is related to the notion that the neural membrane is a “leaky integrator,” which means that more charge is needed to compensate membrane leakage with longer than shorter pulse durations in order to excite a neural response (Parkins & Colombo, 1987; Shepherd & Javel, 1999).…”

Section: Discussionmentioning

confidence: 99%

The Perception of Ramped Pulse Shapes in Cochlear Implant Users

et al. 2021

View full text Add to dashboard Cite

The electric stimulation provided by current cochlear implants (CI) is not power efficient. One underlying problem is the poor efficiency by which information from electric pulses is transformed into auditory nerve responses. A novel stimulation paradigm using ramped pulse shapes has recently been proposed to remedy this inefficiency. The primary motivation is a better biophysical fit to spiral ganglion neurons with ramped pulses compared to the rectangular pulses used in most contemporary CIs. Here, we tested the hypotheses that ramped pulses provide more efficient stimulation compared to rectangular pulses and that a rising ramp is more efficient than a declining ramp. Rectangular, rising ramped and declining ramped pulse shapes were compared in terms of charge efficiency and discriminability, and threshold variability in seven CI listeners. The tasks included single-channel threshold detection, loudness-balancing, discrimination of pulse shapes, and threshold measurement across the electrode array. Results showed that reduced charge, but increased peak current amplitudes, was required at threshold and most comfortable levels with ramped pulses relative to rectangular pulses. Furthermore, only one subject could reliably discriminate between equally-loud ramped and rectangular pulses, suggesting variations in neural activation patterns between pulse shapes in that participant. No significant difference was found between rising and declining ramped pulses across all tests. In summary, the present findings show some benefits of charge efficiency with ramped pulses relative to rectangular pulses, that the direction of a ramped slope is of less importance, and that most participants could not perceive a difference between pulse shapes.

show abstract

Nonlinear auditory models yield new insights into representations of vowels

Cited by 18 publications

References 57 publications

Dissecting neural computations of the human auditory pathway using deep neural networks for speech

Dissecting neural computations of the human auditory pathway using deep neural networks for speech

Representations of fricatives in sub-cortical model responses: comparisons with human consonant perception

The Perception of Ramped Pulse Shapes in Cochlear Implant Users

Contact Info

Product

Resources

About