结合高斯混合模型和VOT特征的音素发音错误检测

结合高斯混合模型(GMM)和嗓音起始时间(VOT)特征的普通话音素发音错误检测,提出了一种结合语音声道特征信息和音源特征信息的发音错误检测方法。其中GMM用于反映声道特征信息的MFCC参数的建模与评测,并直接对大部分音素的发音质量直接进行错误检测。对于少数通过MFCC参数和GMM难于检测区分的辅音音素,则通过反映VOT信息的音源特征参数进行区分。实验表明,该方法在训练数据有限的情况下取得了较好的性能,非常适合用于聋人语言康复的计算机辅助训练。

Phonetic mispronunciation detection based on GMM and VOT

Based on Gaussian mixture model (GMM) and voice onset time (VOT), a novel Mandarin phonetic mispronunciation detection approach was proposed which combines the spectral feature and the prosodic feature. GMM was used to model the Mel-frequency cepstral coefficients (MFCCs) and assess the pronunciation of most phonemes. For some consonants which are difficult to distinguish, prosodic features reflecting the VOT were extracted and used directly. Mandarin phonetic mispronunciation detection experiments on limited data showed the significant improvement.