虚拟式语音信号实时采集与识别平台

Virtual Real-time Speech Signal Acquisition and Recognition Platform

提出了基于LabVIEW和MATLAB混合编程的虚拟式语音信号采集、分析与识别平台设计方案;该平台通过PC机声卡完成对语音信号的实时采集和回放,并对语音信号进行时频分析、端点检测和倒谱分析,进而实现语音识别;为提高识别的正确性和灵活性,提出孤立字列举识别算法和词库编码连续语音识别算法;结果表明,两种算法都能有效识别同音字和读音相似的字;其中,词库编码识别算法不仅效率高,还具有一定的纠错能力,更具有应用价值。

A design of speech acquisition, analysis and recognition platform was proposed based on mixed programming of LabVIEW and MATLAB. It realized real--time speech acquisition and playback through sound card. The features of speech signal in time and frequency domains were analyzed, such as endpoint and MFCC. In order to improve the correctness and flexibility of recognition, two simple and effective speech recognition methods were proposed： one is listing identification for isolated words recognition, the other is word--coding identification for continuous speech recognition. The results proved that the proposed algorithms can recognize two homophones and similar words effectively. Between them, word--coding identification algorithm not only perform efficiently, but also has certain error correction ability and is of great application value.