Self-Adjustable Speech Enhancement and Recognition System
Tomoko Kawase, Manabu Okamoto, Takaaki Fukutomi, Yamato Takahashi, Ryota Masuda, Takayuki Ootake · 2019
We have developed a self-adjustable speech enhancement and recognition (SSER) system to make automatic speech recognition (ASR) robust to variation of acoustic conditions. The SSER system enhances speech components in observed signals, switching parameter values in accordance with the acoustic conditions. Candidates for the parameter value are automatically generated using a real-coded genetic algorithm (GA) in advance. The experimental results show that the performance of the parameter-set values is improved by the proposed method.