Discriminative multi-resolution sub-band and segmental phonetic model combination
Discriminative multi-resolution sub-band and segmental phonetic model combination
复制标题
判别性多分辨率子带和分段语音模型组合
DOI：
--
复制
发表时间：
2000
期刊：
影响因子：
--
通讯作者：
S. Vaseghi
中科院分区：
文献类型：
--
作者：
P. McCourt;N. Harte;S. Vaseghi
提出了一种用于多分辨率子带HALF和混合分段语音模型的联合判别框架，该框架使用训练到最小分类误差标准的类相关权重来组合独立的似然分数。这成功地将语音分类任务的性能扩展到76.3%，而全频带HMM得分为69.2%。
A joint discriminative framework for multi-resolution sub-band HMMs and a hybrid segmental phonetic model is presented which combines independent likelihood scores using class dependent weightings trained to a minimum classification error criteria. This successfully extends the performance on a phonetic classification task to 76.3% compared to a full-band HMM score of 69.2%.