12th Speech in Noise Workshop, 9-10 January 2020, Toulouse, FR

The influence of a physiologically inspired complex compression scheme on perceived listening effort for speech in noise

Saskia M. Waechter(a), Vinzenz H. Schönfelder, Sarah Voice, Nicholas R. Clark
Mimi Hearing Technologies GmbH, Research Department, Berlin, Germany

(a) Presenting

Objective: The primary goal of this study was to assess whether the required listening effort for speech recognition can be decreased by processing a clean speech signal with a complex compression scheme consisting of an instantaneous feed-forward and delayed feedback component mimicking the early stages of the healthy human auditory system ("Mimi-processing"). Listening effort measures were compared between processed and unprocessed sentences. A global equal-RMS constraint was imposed to avoid the influence of level-boost.

Methods: Perceived listening effort was assessed for 30 participants between the ages of 21 to 60 years old (mean = 32.5 ± 10.7 years SD) with the ACALES procedure (Krueger et al., 2017). Participants had average PTA4s of 9.4 dBHL (SD= 5.5 dBHL) in their better ear. The ACALES method employs a rating scale which is applied in an adaptive procedure to measure perceived listening effort for a wide range of (individualised) SNRs without resulting in ceiling effects. Participants rated their effort from 1=’(almost) no effort’ to 13=’Extremely effortful’ or 14=’Only Noise’. Sounds were presented binaurally via Etymotic ER-1 insert earphones. The cohort was divided into three groups for which three different noise types were assessed, namely speech-shaped noise (SSN), multi-talker babble (MTB) and Cafeteria noise.

For each condition and participant, a two-slope function was fitted to the data points and the SNR-distance between the fitted functions of two different listening conditions at equal ratings is the measure of interest. The mean SNR-distance across ratings was calculated per participant and provides a value for how much the SNR can differ between two conditions and yet provide equal average effort ratings.

Results: SNR-differences [dB] between processed and unprocessed stimuli were significantly different from zero (p<0.001) with mean Mimi-processing benefits of 2.55 dB (SSN), 2.31 dB (MTB noise) and 2.22 dB (Cafeteria noise). This means that after Mimi-processing, speech stimuli with SNRs reduced by 2.22dB - 2.55 dB (noise dependent) are rated at equal listening effort by the average listener compared to unprocessed stimuli.

Conclusions: These results indicate that the Mimi-processing algorithm can decrease the perceived listening effort for speech presented in noise. This work provides a promising foundation upon which further improvements of the processing parameters may be implemented to increase speech intelligibility in noise.

Last modified 2020-01-06 19:23:55