Speech synthesis algorithm for Thai cochlear implants

Abstract

Enhancement of speech perception is a crucial aspect for cochlear implant (CI) technology. In a tonal language such as Thai, with segments (consonants and vowels) and supra-segments (tones), many crucial acoustic cues are to be taken into account for speech processing strategy, i.e., amplitude envelopes and temporal fine structure. This paper presents a new speech synthesis algorithm for CI, which combines an amplitude envelope extraction by the continuous interleaved sampling (CIS) strategy and a temporal fine structure extraction to construct carrier signals, and to use them for amplitude modulation based on each band's envelope. Synthesized initial phonemes using three different algorithms (Fu et al., Chen and Zhang, and the proposed algorithm) with six-band speech processor were evaluated by eight native Thais with normal hearing. The results show that the proposed algorithm produces the most significant improvement in speech intelligibility of the initial phonemes compared to the other algorithms. This suggests a possibility of using the proposed algorithm for CI devices, specifically designed for tonal languages such as Thai, to improve speech perception in CI patients.