Search

This 7-month-long project quantifies the role of brain rhythms in speech perception by measuring intelligibility of spoken sentences with judiciously manipulated changes in syllabic rhythm. Speech was time-compressed by a factor of three, resulting in a signal with a syllabic rate three times faster than the original and with poor intelligibility (< 50% words correct). An artificial "syllabic" rate was then introduced by segmenting the time-compressed speech signal into consecutive 40-ms intervals, each followed by a variable interval of silence. The parameters of interest were the length of the silent intervals inserted (ranging between 0-160 ms) and whether the intervals were equal in length (i.e., periodic) or not (i.e., aperiodic). The resulting performance curve is U-shaped, with best intelligibility measured at silence interval of 80 ms inserted periodically. This is also the condition in which there is a significant difference in intelligibility between periodic and aperiodic insertion (the error rate of the latter is nearly twice as high). The U-shaped performance curve may reflect the operation of cortical rhythms. Optimum intelligibility is associated with waveform-energy fluctuations in the core of the theta range of neural oscillations (3-8 Hz), which is also the core range of syllabic rate in naturally spoken utterances. Poor intelligibility may reflect the mismatch between waveform-energy fluctuations and theta rhythms in the brain.

We developed a computational model of diphone perception based on salient properties of peripheral and central auditory processing. The model comprises an efferent-inspired closed-loop model of the auditory periphery (PAM) connected to a template-matching circuit (TMC). Robustness against background noise is provided principally by the signal processing performed by the PAM, while insensitivity to time-scale variations is provided by properties of the TMC. The PAM parameters were determined in isolation from the TMC. This was achieved by analyzing confusion patterns generated in a paradigm with a minimal cognitive load (the binary Diagnostic Rhyme Test [DRT], with synthetic speech stimuli to restrict phonemic variation). Originally, we intended to test the model by quantifying its ability to predict human performance in perceiving naturally spoken speech in the presence of noise, in two separate tasks: (1) diphone discrimination of minimal word-pairs (Voiers' DRT), and (2) phone identification of schwa-CVC tokens. Eventually, the model was evaluated using synthetic speech material.

The overall goal of the STTR program is to formulate a template-matching operation, with perception-related rules of integration over time and frequency at its core, in the context of human perception of degraded speech. In particular, we aim at developing models of auditory processing capable of predicting consonant confusion by normally-hearing listeners, under a variety of acoustic distortions A prerequisite is to formulate the signal processing principles realized by the auditor periphery in providing the observed graceful degredation of human performance in noise. In the nominal four quarters of Phase I we have focused on the role of the descending auditory pathway in regulating the operating point of the cochlea, resulting in auditory nerve (AN) representation of speech sounds that are less sensitive to changes in sustained background noise. A closed-loop model of the auditory periphery, with efferent-inspired feedback, ha been implemented that produces spectrograms of noisy speech that are more consistent with spectrograms of speech in quiet than are spectrograms produced by open-loop models of the auditory periphery., The original document contains color images.

This report summarizes work that has been completed in Phase I of the STTR program. We have implemented a closed-loop model of the auditory periphery with an efferent-inspired feedback and have quantitatively demonstrated its ability to produce spectrograms of noisy speech samples that are far more consistent with spectrograms of speech in quiet than are spectrograms produced by an open-loop model of the auditory periphery. This increase in performance in noise and increased robustness mimics the general observed behavior of humans. Whether this model of auditory periphery preserves phonetic information in patterns that follow psychophysical patterns will be rigorously inspected during Phase II, where the central part of the proposal, i.e. the formulation of a perception-based distance measure, will be established.

This is a final report for a stand-alone grant supporting the first 9 months of a 4-year research program entitled "Auditory peripheral processing of degraded speech". The underlying thesis is that the auditory periphery contributes to the robust performance of humans in speech reception in noise through a concerted contribution of the efferent feedback system (which stabilizes the operating point of the cochlea) and post-auditory-nerve functions (such as acoustic-edge detectors). The 4-year program includes psychophysical experiments aimed at documenting the role of these two peripheral components; developing a computational model; and using the emerging model as a front end to an automatic speech recognition system. The proposed timetable allocates the first 12 months for psychophysical experiments. This report describes efforts to assess the role of the MOC efferent system in speech reception in the presence of sustained background noise, by using a test of initial consonant reception (the Diagnostic Rhyme Test) performed by subjects with normal hearing. Activation of selected parts of the efferent system was attempted by presenting speech and noise in various configurations (gated/continuous, monaural/binaural). Initial results support the hypothesis of a significant efferent contribution to initial phone discrimination in noise.