| Summary: | 碩士 === 國立成功大學 === 電機工程學系碩博士班 === 91 === In this thesis, an environmental sound recognition system based on MPEG-7 features (centroid, spread, and flatness [1]) and its corresponding VLSI architectures are proposed. Traditional sound recognizer utilizes decision-tree based method and causes a problem where the parameter is not generalized [2~5]. The HMM based sound recognizer has been introduced by [8] to resolve this drawback. However, it adopts spectrum parameter and will result in high dimensional feature vectors. This thesis successfully solves the shortcoming by taking the basis extraction. The recognition rate is about 82% while only spectrogram is adopted as the parameter. The improved recognition rate is about 95% while above three mentioned MPEG-7 audio features are regarded as the parameters in our environmental sound recognizer.
Moreover, related VLSI architectures for this sound recognition system are also proposed. The first one is the feature extraction module. The most complicated computations in the module are the division and nth-root operations. We utilize the CORDIC method to devise a divider. For the nth-root operation, a specific circuit is designed in accordance with the Brahmagupta iteration algorithm. For the Viterbi algorithm, a dedicated hardware architecture is also presented. This architecture is designed based on the 4-step fully Viterbi algorithm. This speed-up of this module is also ascribed to the fully pipeline systolic array architecture.
|
|---|
