Correlation distance skip connection denoising autoencoder (CDSK-DAE) for speech feature enhancement

Abstract

Performance of learning based Automatic Speech Recognition (ASR) is susceptible to noise, especially when it is introduced in the testing data while not presented in the training data. This work focuses on a feature enhancement for noise robust end-to-end ASR system by introducing a novel variant of denoising autoencoder (DAE). The proposed method uses skip connections in both encoder and decoder sides by passing speech information of the target frame from input to the model. It also uses a new objective function in training model that uses a correlation distance measure in penalty terms by measuring dependency of the latent target features and the model (latent features and enhanced features obtained from the DAE). Performance of the proposed method was compared against a conventional model and a state of the art model under both seen and unseen noisy environments of 7 different types of background noise with different SNR levels (0, 5, 10 and 20 dB). The proposed method also is tested using linear and non-linear penalty terms as well, where, they both show an improvement on the overall average WER under noisy conditions both seen and unseen in comparison to the state-of-the-art model. (Hanseok Ko) constructed of word sequence probabilities given a phone sequence [1].Over the last decade, there has been a dramatic shift in the ASR community due to the explosion of deep learning. Although the idea of deep layered network has been around over many decades, the sudden embrace of the method by the ASR community can be attributed to availability of both computational power and large public datasets that would allow training of deep layered networks possible. Effectiveness of deep networks has been such that traditionally handcrafted feature based methods, such as GMM, was mostly replaced by the new paradigm. The approach yielded a significant improvement in performance of ASR systems [2][3][4]. Application of deep network to other areas such as lexicon and language followed for improved