Method for enhancing robustness of voice recognition system

Abstract

The invention provides a method for enhancing the robustness of a voice recognition system. The method comprises the following steps of: updating a long term average value of power spectrum of a voice signal and a minimum valve of the long term average value of the power spectrum of the voice signal according to transient power spectrum of a current signal segment; calculating the signal-to-noise ratio, of the voice signal, in a first frequency domain according to the minimum value of the long term average value of the power spectrum; judging whether an effective voice exists in different frequency distribution areas according to the signal-to-noise ratio in the first frequency domain; estimating transient power spectrum of a first noise based on the judgment; updating the long term average value of the power spectrum of a noise and the minimum value of the long term average value of the power spectrum of the noise according to the estimated power spectrum of the first noise; re-estimating the signal-to-noise ratio, of the voice signal, in a second frequency domain according to the updated minimum value of the long term average value of the power spectrum of the noise; estimating the probability of the existence of the voice in a frequency domain according to the secondarily-estimated signal-to-noise ratio in the second frequency domain; and estimating the power spectrum of a second noise based on the probability of the existence of the voice.

Description

technical field The present invention relates to automatic speech recognition technology, in particular, to a method for enhancing the robust performance of a speech recognition system in a noisy environment. Background technique Nowadays, speech recognition technology is widely used. For example, a call center uses a large-scale speech recognition system to recognize users' voice content and automatically answer users' questions through speech analysis technology. Additionally, Internet services can be accessed by voice with the help of a voice recognition engine. For example, some Internet search service providers also launched voice search services based on mobile phone platforms, which largely solved the problem of inconvenient input of text when users browse the web on mobile phones, and improved user satisfaction. As another example, a large number of electronic toys are equipped with embedded speech recognition engines, so that they can interact with players through...

AI Technical Summary

Problems solved by technology

However, general solutions are still to be developed

Another problem is that the current speech recognition schemes all have different disadvantages, for example, some schemes need to carry a huge amount of computation, while others require a large noise training library

Images