Speech feature extraction method and reconfigurable speech feature extraction device in multi-noise scene

Abstract

The invention discloses a speech feature extraction method and a reconfigurable speech feature extraction device in a multi-noise scene, belonging to the technical field of speech recognition. The present invention combines the characteristics of low power consumption of low-pass filter speech extraction and the high accuracy of Mel filter speech feature extraction, and dynamically selects speech feature extraction according to the analysis and judgment results of the bottom noise threshold and the output results of the low-pass filter and neural network In this way, the device switches the speech feature extraction channel through the reconfigurable feature extraction function configuration module. Under the condition that there is no speech in the external environment or there is speech but the signal-to-noise ratio is high, the low-pass filter is used to extract the speech features and the recognition of the neural network is performed, while the Mel is used when the signal-to-noise ratio is low and there is speech input. The filter performs speech feature extraction to reduce the overall power consumption of speech feature extraction.

Description

technical field [0001] The invention discloses a speech feature extraction method and a reconfigurable speech feature extraction device in a multi-noise scene, relates to artificial intelligence neural network technology, and belongs to the technical field of speech recognition. Background technique [0002] At present, with the development of speech recognition technology, digital equipment and multimedia technology, speech endpoint detection technology has been well developed. Speech endpoint detection is a technology for detecting speech segments in continuous signals. Speech endpoint detection can be used with automatic speech recognition systems. , voiceprint recognition system, voice endpoint detection can provide accurate and effective voice endpoints, so voice endpoint detection is a very important module. [0003] To detect voice endpoints at all times, the voice endpoint detection module must be turned on all the time. At this time, the power consumption of the ent...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to address the shortcomings of the above-mentioned background technology and provide a speech feature extraction method and a reconfigurable speech feature extraction device in a noisy scene, which can dynamically select the speech feature extraction method to reduce power consumption while maintaining accuracy , which solves the technical problems of low power consumption and low precision or high power consumption and high precision of the existing voice endpoint detection module

Images