A Novel Speech Processing Algorithm Based on Harmonicity Cues in Cochlear Implant
Jian Wang,Yousheng Chen,Zongping Zhang,Yan Chen,Weifeng Zhang
DOI: https://doi.org/10.1063/1.4993000
2017-01-01
Abstract:This paper proposed a novel speech processing algorithm in cochlear implant, which used harmonicity cues to enhance tonal information in Mandarin Chinese speech recognition. The input speech was filtered by a 4-channel band-pass filter bank. The frequency ranges for the four bands were: 300-621, 621-1285, 1285-2657, and 2657-5499 Hz. In each pass band, temporal envelope and periodicity cues (TEPCs) below 400 Hz were extracted by full wave rectification and low-pass filtering. The TEPCs were modulated by a sinusoidal carrier, the frequency of which was fundamental frequency (F0) and its harmonics most close to the center frequency of each band. Signals from each band were combined together to obtain an output speech. Mandarin tone, word, and sentence recognition in quiet listening conditions were tested for the extensively used continuous interleaved sampling (CIS) strategy and the novel F0-harmonic algorithm. Results found that the F0-harmonic algorithm performed consistently better than CIS strategy in Mandarin tone, word, and sentence recognition. In addition, sentence recognition rate was higher than word recognition rate, as a result of contextual information in the sentence. Moreover, tone 3 and 4 performed better than tone 1 and tone 2, due to the easily identified features of the former. In conclusion, the F0-harmonic algorithm could enhance tonal information in cochlear implant speech processing due to the use of harmonicity cues, thereby improving Mandarin tone, word, and sentence recognition. Further study will focus on the test of the F0-harmonic algorithm in noisy listening conditions.