Noise removing method in intelligent calling system based on CNN

Abstract

The invention discloses a noise removing method in an intelligent calling system based on a CNN. The method comprises the following steps: carrying out slicing processing, normalization and framing preprocessing on a telephone signal; intercepting a slice signal after framing; extracting the Mel frequency spectrum of the intercepted signal; inputting the extracted Mel frequency spectrum into a neural network model for model training, wherein the trained classification model is used as a noise classification model; slicing the newly added telephone signal and dividing the telephone signal intoodd segments; carrying out normalization and framing preprocessing on the slice signals; respectively intercepting each section of slice signal after framing; extracting the Mel frequency spectrum ofeach section of intercepted signal; and inputting the extracted Mel frequency spectrum noise into the noise classification model. The method has the advantages that whether the signals are human voiceor noise is identified through the classification model based on the CNN, a large number of noise signals in the telephone signals can be removed, the error rate that the signals are sent to ASR to be translated into characters is reduced, and audio distortion caused by excessive attenuation is avoided.

Description

Technical field [0001] The present invention relates to the technical field of audio processing, and in particular, to a noise removal method in a CNN-based intelligent call system. Background technique [0002] In the existing intelligent call system, the telephone signal will be intercepted by VAD, and then sent to ASR to be converted into text. Due to the complexity of the background, there are a lot of noise fragments. The usual processing method is to use noise suppression methods to filter the signal before the signal is intercepted. It is mainly based on the frequency distribution of the signal to estimate the noise. Commonly used algorithms include adaptive filters, spectral subtraction, and Wiener filtering. The adaptive filter uses the filter parameters obtained at the previous moment to automatically adjust the current filter parameters to adapt to the statistical characteristics of the random changes of the signal and noise, thereby achieving noise filtering; spectra...

AI Technical Summary

Problems solved by technology

These noise suppression methods can only filter part of the noise, but cannot completely remove the intercepted noise segment, and as the signal-to-noise ratio in the telephone signal decreases, the noise reduction effect becomes worse, and there will be some periods of time due to excessive noise. Audio distortion caused by attenuation

Images