Deep Neural Networks (DNNs)

Deep Neural Networks (DNNs)

Deep Neural Networks (DNNs) are extremely powerful in performing machine learning tasks including image classification, speech recognition, and speech coding . However, training DNNs is computationally difficult. In particular, fine tuning DNNs requires stochastic gradient descent, which is unusually difficult to parallelize across machines. This lack of parallelism makes learning at large scale practically impossible.

The goal of the research described in this paper is a scalable method for buidling deep classification architectures. Since the basic learning algorithm of DSN is convex, it can also be called deep convex network. The main theoretical motivation of this work is the desire to learn complex functions from large data sets with parallel learning algorithm.

In recent years, single-channel speech enhancement has attracted a considerable amount of research attention because of the growing challenges in many important real-world applications. Hearing aids design and robust speech recognition. The goal of speech enhancement is to improve the intelligibility and quality of a noisy speech signal degraded in adverse conditions. However, the performance of speech enhancement in real acoustic environments is not always satisfactory. Numerous speech enhancement methods were developed over the past several decades.

Deep Neural Networks (DNNs)

A common problem usually encountered in these conventional methods is that the resulting enhanced speech often suffers from an annoying artifact called “musical noise”. Another notable work was the minimum mean-square error (MMSE) estimator introduced by Ephraim and Malah; their MMSE log-spectral amplitude estimator could result in much lower residual noise without further affecting the speech quality. An optimally-modified log-spectral amplitude (OM-LSA) speech estimator and a minima controlled recursive averaging (MCRA). Although these traditional MMSE-based methods are able to yield lower musical noise. A trade-off in reducing speech distortion and residual noise needs to be made due to the sophisticated statistical properties of the interactions between speech and noise signals.

Leave a Comment