Effect of auditory efferent time-constant duration on speech recognition in noise

Abstract

The human auditory efferent system may play a role in improving speech-in-noise recognition with an associated range of time constants. Computational auditory models with efferent-inspired feedback demonstrate improved speech-in-noise recognition with long efferent time constants (2000 ms). This study used a similar model plus an Automatic Speech Recognition (ASR) system to investigate the role of shorter time constants. ASR speech recognition in noise improved with efferent feedback (compared to no-efferent feedback) for both short and long efferent time constants. For some signal-to-noise ratios, speech recognition in noise improved as efferent time constants were increased from 118 to 2000 ms.

Optimizing Speech Recognition Using a Computational Model of Human Hearing: Effect of Noise Type and Efferent Time Constants

“…The general trend for an increase in speech recognition scores in the absence of efferent activation with increasing SNR is similar to that reported by [16] and [20] for SNR values up to 20 dB. Overall, for SNRs of 0 to 20 dB, there was an improvement in speech recognition with efferent activation, compared to when there was no efferent activation within the model.…”

“…In Experiment 2 of the current study, speech recognition in noise was also measured (using the same model settings) in AM pink noise using AM rates spread across 1-14 Hz (comparable to envelope fluctuation rates in speech). The results will allow for a direct comparison with Experiment 1 (which used UM pink noise), as well as comparison with earlier studies (measuring speech recognition in speech-shaped/babble noise) using auditory models with longer efferent time constants [14]- [16], [20].…”

“…In Experiment 1 of the current study, speech recognition in noise was measured in UM pink noise with time constants ranging from 50 to 2000 ms. The results will allow for a direct comparison with earlier studies measuring speech recognition in UM pink noise, using an auditory model with a single long efferent time constant of 2000 ms [14], [16] or time constants within a range of 118-2000 ms [20]. In Experiment 2 of the current study, speech recognition in noise was also measured (using the same model settings) in AM pink noise using AM rates spread across 1-14 Hz (comparable to envelope fluctuation rates in speech).…”

“…In the present study, the efferent decay time constants tested within the model were: 50, 70, 86, and 100 ms, in order to make a comparison with the shorter range of human auditory efferent time constants recorded using OAEs [18]; 118 ms, an efferent decay constant reported by [28] using human psychoacoustical measures; 200 ms, 450 ms, and 1000 ms which are within the range of slow and medium human efferent decay constants recorded using OAEs [18]; and 2000 ms, in order to make a direct comparison with two previous studies [14], [16]. UM pink noise was used in order to make a comparison with earlier findings [14], [16], [20], and AM pink noise was used in order to investigate the interaction of noise modulation with efferent time constants. UM pink noise or AM pink noise will hereafter be referred to as UM noise or AM noise.…”

“…Previous studies using an auditory model [13], have used only long efferent time constants (e.g., 2000 ms), in order to study speech recognition in pink noise [14], [16]. In order to investigate the role of auditory efferent activation across the range of time constants recorded in humans [18], a previous study by the authors [20] investigated speech recognition in unmodulated (UM) pink noise with an auditory model [13] using shorter time constants (118, 200, 450, 1000 ms), as well as the longer time constant of 2000 ms originally used in earlier studies. A range of SNRs was used in the study (−10, −5, 0, 5, 7, 10, 12, 15 and 20 dB).…”