A speech recognition-based iterative denoising method and chip

Abstract

The invention discloses an iterative denoising method and chip based on speech recognition, comprising: step 1: determining a target speech signal and its target confidence value; step 2: selecting noise data matching the target confidence value from a noise database, And control the unmarked voiced frame in the noise data and the target voice signal to participate in pre-denoising processing; Step 3: judge whether the pre-denoising processing result is greater than the predetermined threshold, if yes, enter step 4, otherwise enter step 5; step 4: mark the voiced frame corresponding to the pre-denoising processing result as a denoised voiced frame in the target speech signal; Step 5: judge the confidence value of the pre-denoising processing result and the target confidence Whether the absolute value of the value difference is less than a confidence threshold, if so, the voiced frame corresponding to the pre-denoising processing result is marked as a denoised voiced frame in the target speech signal; otherwise, the target confidence value is adjusted , and then return to step 2.

Description

technical field [0001] The invention belongs to the technical field of robots, and in particular relates to a speech recognition-based iterative denoising method and a chip. Background technique [0002] Existing mobile robots (such as mechatronics equipment such as sweeping robots, window cleaning robots, and floor mopping robots among cleaning robots) will generate noise during work. The voice signal is picked up by voice, but when the voice signal from the user is picked up, the noise generated during the robot's work is usually picked up by voice, so that the voice signal picked up by the device is mixed with a lot of external noise , the corresponding speech recognition accuracy is not high, which will seriously affect the robot's recognition of external speech (effective signal), and make logical decisions based on speech interpretation (such as performing relevant path planning). [0003] In the prior art, the method of denoising at the front end of the voice signal ...

AI Technical Summary

Problems solved by technology

[0002] Existing mobile robots (such as mechatronics equipment such as sweeping robots, window cleaning robots, and floor mopping robots among cleaning robots) will generate noise during work. The voice signal is picked up by voice, but when the voice signal from the user is picked up, the noise generated during the robot's work is usually picked up by voice, so that the voice signal picked up by the device is mixed with a lot of external noise , the corresponding voice recognition accuracy is not high, which will seriously affect the robot's recognition of external voices (effective signals), and make logical decisions based on voice interpretation (such as performing relevant path planning)

[0003] In the prior art, the method of denoising at the front end of the voice signal is generally to select the appropriate voice signal according to the result of the voice signal classification, and suppress the voice signal that does not meet the requirements, but the method for the voice signal classification is more complicated, not only The noise reduction is not complete, and the efficiency of speech recognition is not high, there are always remaining speech frames that have not been processed, which affects the effect of speech recognition

Images