ROBUST RECOGNITION OF SMALL -VOCABULARY TELEPHONE - QUALITY SPEECH

摘要: Considerable progress has been made in the field of automatic speech recognition recent years, especially for high-quality (full bandwidth and noise-free) speech. However, good accuracy is difficult to achieve when incoming passed through a telephone channel. At same time, task over lines growing importance, as number applications spoken language processing involving increases every day. The paper presents our work on developing robust speaker-independent isolated-spoken word system based hybrid approach (classic – artificial neural network). A experiments are described compared order evaluate different analysis techniques that best suited telephone-speech task. In particular, we address use RASTA (i.e., filtering temporal trajectories parameters) increasing accuracy. Also, propose method adaptive filter theory producing simulated data starting from clean databases.