Influence of noise reduction algorithms on phonemes recognition in the case of Binaural Cochlear Implant coding

French phonemes perception in noisy conditions, in the case of a Binaural Cochlear Implant (BCI) coding, is seen in the present study. In the current work, the action of binaural noise reduction algorithms is investigated, through the use of a vocoder simulation with normal hearing listeners. Three binaural noise reduction algorithms, used in classical hearing aids, have been considered: beamformer, Doerbecker algo- rithm combined with Ephraim and Malah noise estimator and Doerbecker algorithm combined with Scalart noise estimator. Then a cochlear implant (CI) coding (bins grouped into frequency bands) transformed the signal at the end of the processing chain. Also, a percentage of the input signal was "re-injected" (added) before CI coding. Twenty-six normal hearing subjects participated in the experiment and they listened to sessions including 3 signal-to-noise ratios, 3 re-injection coefficients; they evaluated the coded signal (phoneme recognition). Then, a noise was added to jam the signal. The noise came from five different noise angles and the speech was issued from the front (zero deg azimuth). Altogether, experimental sessions tested 150 conditions. Best results were obtained using the beamformer algorithm. Doerbecker with Ephraim and Malah estimator led to good results; this strategy was more efficient than the Doerbecker with Scalart estimator. Results were more sensitive to the speech processing strategy than to the noise angle. Re-injection of the input signal improved the recognition. In this BCI coding environment, noise reduction algorithms led to an improvement of 20% in phoneme recognition.