To solve the problem of speech feature extraction in noisy environment, an unsupervised speech denoising method based on deep neural network is proposed in this paper. The deep neural network is used to estimate the noisy speech, and the real-time received speech signal is transformed in S-domain. The power spectrum characteristic information is analyzed, and the restricted Boltzmann machine is used to train and tune the real-time speech information unsupervised, decode the model trained in the off-line stage, and get the estimation of logarithmic power spectrum of speech. Then synthesize the subjective audible speech waveform file with the phase of noisy speech to improve the anti-noise ability in speech feature extraction.
Paper
Full text
Unsupervised Speech Denoising Method Based on Deep Neural Network
Semantic Scholar · Computer Science · 2018
Abstract
To solve the problem of speech feature extraction in noisy environment, an unsupervised speech denoising method based on deep neural network is proposed in this paper. The deep neural network is used to estimate the noisy speech, and the real-time received speech signal is transformed in S-domain. The power spectrum characteristic information is analyzed, and the restricted Boltzmann machine is used to train and tune the real-time speech information unsupervised, decode the model trained in the off-line stage, and get the estimation of logarithmic power spectrum of speech. Then synthesize the subjective audible speech waveform file with the phase of noisy speech to improve the anti-noise ability in speech feature extraction.