基于空间相关性变换的声学模型训练

苏腾荣; 吴及; 王作英

doi:10.3724/SP.J.1146.2009.00343

基于空间相关性变换的声学模型训练

doi: 10.3724/SP.J.1146.2009.00343

计量
- 文章访问数: 3549
- HTML全文浏览量: 115
- PDF下载量: 841
- 被引次数: 0
出版历程
- 收稿日期: 2009-03-16
- 修回日期: 2009-08-17
- 刊出日期: 2010-04-19

Acoustic Model Training Based on Spatial Correlation Transformation

摘要

摘要: 为了在语音识别中增强对不同语音单元之间的相关性的利用，该文基于空间相关性变换(Spatial Correlation Transformation，SCT)框架，提出一种新的模型训练算法，在说话人无关模型的训练中利用训练数据中的空间相关性进行模型参数重估。该算法对所有训练数据进行空间相关性变换，削弱数据间的空间相关性，使重估的模型更不依赖训练数据，以改善模型的性能。实验表明，基于空间相关性变换框架的模型训练方法与基于该框架的特征变换方法相结合，使系统的平均错误率相对基线系统下降了18%。
- 语音识别; 空间相关性; 特征变换; 模型训练
Abstract: In order to enhance the utilization of the correlation between different acoustic units in speech recognition, a novel model training approach based on the Spatial Correlation Transformation (SCT) framework is proposed in this paper, in which the speaker-independent model parameters are re-estimated using the spatial correlation information in the training data. In this algorithm, SCT is applied to all training data, to decrease the correlation among the training data, make the model re-estimated less dependent on the training data, and then improve the performance of the model. Experiments show that the combination of SCT-based model training and SCT-based feature transformation achieves a relative reduction of 18% of average syllable error rate compared to the baseline system.

HTML全文

参考文献(1)

Leggetter C J and Woodland P C. Maximum likelihoodlinear regression for speaker adaptation of continuousdensity hidden markov models[J].Computer Speech andLanguage.1995, 9(2):171-185[2]Kuhn R, Junqua J C, and Nguyen P, et al.. Rapid speakeradaptation in eigenvoice space[J].IEEE Transactions on Speechand Audio Processing.2000, 8(6):695-707[3]Anastasakos Tasos, McDonough John, and Makhoul John.Speaker adaptive training: A maximum likelihood approachto speaker normalization. Proceedings of ICASSP, Munich,Germany, 1997: 1043-1046.[4]Sinha R and Gales M J F, et al.. The CU-HTK mandarinbroadcast news transcription system. Proceedings ofICASSP, Toulouse, France, 2006: 1077-1080.[5]Ng Tim, et al.. Progress in the BBN 2007 mandarin speech totext system. Proceedings of ICASSP, Las Vegas, USA, 2008:1537-1540.Su Teng-rong, Wu Ji, and Wang Zuo-ying. Spatialcorrelation transformation based on minimum covariance.Proceedings of ICASSP, Las Vegas, USA, 2008: 4697-4700.[6]苏腾荣, 吴及, 王作英. 空间相关性变换及其在语音识别中的应用. 清华大学学报( 自然科学版), 2009, 49(10):1655-1659.Su Teng-rong, Wu Ji, and Wang Zuo-ying. Spatial correlationtransformation and its application in speech recognition.Journal of Tsinghua University (Science and Technology),2009, 49(10): 1655-1659.[7]王作英, 肖熙. 基于段长分布的HMM语音识别模型. 电子学报, 2004, 32(1): 46-49.Wang Zuo-ying and Xiao Xi. Duration distribution basedHMM speech recognition models. Acta Electronica Sinica,2004, 32(1): 46-49.

施引文献

资源附件(0)

访问统计