A study on the recognition of low bit-rate encoded speech

Digital speech communications are the future trend in the Internet and mobile phones. The low bit-rate coding of speech signals is the essential requirement in the concern of channel bandwidth and transmission efficiency. The voice-based services will become more attractive to the service providers. Many voice-driven applications require that users must be authorized and able to be identified. Other applications may allow users to retrieve information or perform some kind of transactions by voice. Traditionally, speech recognition systems need to get the waveform of speech signal and convert it into some specific feature parameters. In order to retain enough speech information, the sampling rate as well as the bit count per sample is high. In other words, these speech recognition systems require high bit-rate speech data. In the contrast, speech coding is to compress the speech information at low bit-rate. To recognize the encoded speech should face the problem of limited and inaccurate feature parameters. One may decode the coded information to recover the speech signal. However, the recovered signal can not retain the same waveform as the original one. Further process of this recovered signal does not gain any extra information.