A mandarin speech recognition result evaluation algorithm on mixed words

Speech recognition result is always evaluated by dynamic programming algorithm which can find the best match between recognized results and references. And the match unit could be word, syllable, phone and so on. However, in mandarin speech recognition, this traditional algorithm is not accurate because of the unique characteristics of Chinese word-formation. In order to improve evaluation performance on Chinese, we proposed a mandarin speech recognition evaluation algorithm based on mixed words. Our experiments show that our proposed algorithm can effectively reduce the false error caused by traditional evaluation algorithm.