The Study Of Keyword Recognition In Speech Indexing Application

Keyword spotting is an important area in speech recognition. Its objective is to identify and verify a few specified key-words in continuous speech. Comparing with keyword spotting, continuous speech recognition need more resources and its process speed is slower, and it's more vulnerable to noise. So in the art of state, continuous speech recognition is not suitable in many applications and keyword spotting is preferred. If we greatly advance this technology, it will be a great help to expand speech recognition applications. Apparently information indexing is a very good application.This paper is mainly focus on the system development in keyword spotting and indexing. It provided an excellent keyword spotting engine and a keyword indexing system in the internet. It developed some new methods and algorithms in keyword spotting to achieve a high detection rate and a low false alarm. The main work is as follows:1. The construction of recognition network for recognition engine based on statistical language model. We trained a bigram model with Chinese syllable as the base phoneme using the speech database transcriptions. Then we translate the bigram to the recognition network with filler models and keyword models. The keyword model is composted from syllable model. The keyword recognition is based on the recognition network.2. The training of the syllable recognition element and the triphone model. It is optimized with Chinese recognition element selections, which greatly enhance the differential between recognition elements. Triphone model is used as filler model.3. The paper provided a post process method based on syllable aligning. With the first spotting results, it aligns the recognized keyword in the new syllable network, which is constructed by the keyword, to verify the results.4. The development of audio indexing system named SAS. It can find the audio files which contain the pronunciation of specified keywords on the specified web site, the mp3, rm and wav file formats are acceptable. It is a large vocabulary Chinese keyword spotting and indexing system with an acceptable performance.This work is supported by National Natural Science Foundation of P.R.China(60273059), Zhejiang Provincial Natural Science Foundation for Young Scientist of P.R.China (RC01058), Zhejiang Provincial Natural Science Foundation (M603229) and National Doctoral Subject Foundation (20020335025).