Audio howling suppression method, device and system and neural network training method

Abstract

The invention discloses an audio howling suppression method, device and system and a neural network training method, and the method comprises the steps: obtaining a plurality of continuously collectedfirst audio frames in a time domain; converting the plurality of first audio frames into a plurality of second audio frames respectively, wherein the second audio frames are audio signals in a frequency domain, and each of the plurality of second audio frames comprises a plurality of frequency points; determining howling frequency points with howling signals in the plurality of second audio frames, wherein the howling signals are continuous, single-frequency and high-energy signals; in the second audio frame with the howling signal, removing the audio signal of the howling frequency point; inputting the frequency point signal near the howling frequency point in the plurality of second audio frames into the howling frequency point restoration neural network model, and outputting the restored howling frequency point replacement audio signal to replace the audio signal at the howling frequency point position. Howling signals are suppressed relatively easily and simply, fault of audio signals is avoided, and distortion of the audio signals is reduced.

Description

technical field [0001] The invention relates to the technical field of audio data processing, in particular to an audio howling suppression method, device and system, and a neural network training method. Background technique [0002] Howling is a harsh sound produced by positive feedback between the pickup and the loudspeaker; the inventors have found that howling signals are often single-frequency signals, so in the frequency domain it appears as an abnormality of continuous frames at the same frequency, and also That is, a continuous high-energy signal in the time domain. [0003] Today, power amplifier equipment is used in all aspects of life, collecting voice signals through microphones, and then amplifying and outputting the voice signals through speakers. For example, speeches, concerts or singing in KTV private rooms, etc. In these scenarios, when the microphone and the speaker are very close, the voice signal played by the speaker is collected by the microphone, a...

AI Technical Summary

Problems solved by technology

The adaptive feedback cancellation method system is complex and the filter update is too slow to affect the convergence speed, and the update is too fast to damage the useful signal, and the microphone signal and the feedback signal are strongly correlated, which is difficult to control

Images