PMC在噪声环境下的语音识别中的应用

Application of PMC in Noise Speech Recognition

今天的语音识别正处于由实验室技术走向实用化、产品化的关键时期。然而 ,现有的绝大多数语音识别系统在噪声环境中的性能都不可避免地急剧下降。环境噪声已经成为语音识别技术商品化的一个主要障碍。因此在语音识别技术逐渐走向实用化的过程中 ,噪声语音识别日益成为一个重要的研究领域。遗憾的是 ,由于噪声语音识别问题本身的复杂性 ,至今还没有一种方法可以圆满地解决这一问题。拟从模型补偿方面 ,对噪声环境下的孤立词语音识别进行一些探索。重点研究一个在噪声环境下的语音识别算法——并行模型组合方法 (PMC) ,详细论述了其原理以及在噪声环境下的语音识别中的应用。实验中 ,我们使用汉语的数字语音 ,分别在 3种不同噪声不同信噪比条件下对这一方法进行了识别率测试 ,结果显示 ,该方法有着令人振奋的识别效果。

In the field of speech recognition, we have come to the point where we are concerned with how the technology can be applied to new products and how the technology will change our future. However, the performance of most current speech recognizers is seriously affected when environmental noise occurs during operation. Environmental noise has become one of the major obstacles to the using of speech recognition techniques in practice. Therefore, robust speech recognition has become an important area when speech recognition technology began to be transferred from laboratory to applications field. Due to the complex nature of noisy speech recognition, there has been no single technique that can solve this problem satisfactorily up to now. The purpose of this thesis is to explore isolated word speech recognition in noise environment on model compensating. This paper addresses a kind of method of automatic speech recognition in the presence of interfering noise: Parallel Model Combination(PMC). The theory of PMC method and the using in noisy speech recognition is discussed in detail. And a test with several kinds of noise and different SNR using Chinese digit speech is done. The result shows that PMC is a good method of speech recognition in the presence of interfering noise.