K子空間和時(shí)延自相關(guān)器的英漢音素識(shí)別

提出了用于音素識(shí)別的K子空間和時(shí)延自相關(guān)器神經(jīng)網(wǎng)絡(luò)結(jié)構(gòu)，用將時(shí)延設(shè)計(jì)加入線性自相關(guān)器，以擴(kuò)展音素濾波神經(jīng)網(wǎng)絡(luò)的方法，產(chǎn)生p維子空間，并采用迭代過(guò)程修改劃分，以便捕獲語(yǔ)音信號(hào)中的時(shí)間序列信息。這種帶不分類訓(xùn)練過(guò)程的體系結(jié)構(gòu)提供了一種高識(shí)別性能的方法，沒(méi)有大多數(shù)常規(guī)語(yǔ)音識(shí)別神經(jīng)網(wǎng)絡(luò)所常有的網(wǎng)絡(luò)輸出值不表示候選者似然性的缺陷。通過(guò)英語(yǔ)音素和漢語(yǔ)音素的初步試驗(yàn)，識(shí)別正確率為84.38%，比音素濾波神經(jīng)網(wǎng)絡(luò)方法好。

A neural network architecture, K-subspaces and time-delay auto-associators, is proposed for phoneme recognition. It extends the phoneme filter neural networks approach by adding linear auto-associators to create p-dimension subspace, and an iteration is employed to improve the decision. It is good to capture the time-sequence information in speech signal. The architecture proposed could provide a high recognition performance without traditional neural network’s shortcoming. Some recognition simulations for both English and Chinese phonemes are conducted, and the recognition rate is 84.38% which is better than phoneme filter neural networks approach.