A Surround Sound Error Repair Method in Mobile Network Environment

Abstract

The present invention provides a method for repairing surround sound bit errors in a mobile network environment, using a neural network based on deep learning to learn the nonlinear relationship between the lost frame and its adjacent frame data, and using a method that can characterize the nonlinear perception of the human ear. The logarithmic power spectrum of the characteristic is used as the feature, and the stacked self-encoding model algorithm is used. First, the greedy layer-by-layer unsupervised pre-training is performed for initialization to avoid the trouble of falling into the local optimal solution, and then the supervised learning is used to optimize the loss of data. The prediction is more accurate, using the phase spectrum of the previous frame as the spectrum estimate, and performing the inverse discrete Fourier transform to obtain an accurate estimate of the current lost signal in the time domain. The average error of the surround sound error repair method has dropped by about 25%, and the speech Both the audio and music audio have very good effects. The restored surround sound audio quality is very high, fully retaining the advantages of a strong sense of space, more natural and vivid surround sound, less delay, and good practicability.

Description

technical field [0001] The invention relates to a surround sound error code repair method, in particular to a surround sound error code repair method under a mobile network environment, and belongs to the technical field of audio code error repair methods. Background technique [0002] The rapid development of the mobile Internet has led to the rapid development of entertainment and social networking. As the most convenient and fastest way of interaction, audio signals are in increasing demand and widely used. Different from traditional wired circuit transmission, due to jitter and delay in the transmission of various audio signals in the mobile network, data packet errors and loss will inevitably occur, which seriously affects the user's perception experience. In order to reduce the packet loss rate, some existing technologies will set a jitter buffer at the audio receiving end to ensure the continuity of the received audio data packets to a certain extent and reduce packet...

AI Technical Summary

Problems solved by technology

Although these waveform-based restoration methods are easy to implement, they can cause audio phase mismatch, which seriously affects audio quality at high bit error rates

In the prior art, an audio bit error repair algorithm based on linear prediction and high-order autoregressive models has also appeared, which uses the pitch period to copy the previous frame excitation signal as the current frame excitation signal, or uses the minimum mean square error to recursively recursively correct the lost audio data. Prediction, but these existing methods can produce annoying perceptually severe distortions because the reconstructed excitation signal will be used to reconstruct the lost signal of the next frame

Most of the audio error repair methods in the prior art are aimed at monophonic audio, but there are relatively few studies on the error concealment technology of multi-channel and stereo audio, although some methods in the prior art have considered both data and sound in the channel. However, in actual operation, only one of the intra-channel and inter-channel data is active at the same time, and the synergy between the intra-channel and inter-channel audio data in error recovery is not fully considered. Poor error recovery performance

[0005] On the whole, the existing technology mainly has the following defects: First, the audio phase of the existing waveform-based repair method does not match, which seriously affects the audio quality under high bit error rates; second, the existing technology is based on linear prediction and high-order autoregressive The audio error repair algorithm of the model uses the pitch period to copy the excitation signal of the previous frame as the excitation signal of the current frame, or uses the minimum mean square error to recursively predict the lost audio data, because the reconstructed excitation signal will be used to reconstruct the next frame. A signal with one frame loss will produce annoying perceptual severe distortion; third, most of the audio error repair methods in the prior art are aimed at monophonic audio, while the research on error concealment technology for multi-channel and stereo audio is relatively Less, these methods are suitable for surround sound bit error repair and have almost no repair effect, even play a worse effect, can not reach the purpose of surround sound error bit repair; Fourth, in order to reduce the packet loss rate, the existing technology has A jitter buffer will be set on the audio receiving end to ensure the continuity of the received audio data packets to a certain extent and reduce packet loss, but if the buffer is too large, it will cause a huge audio data delay and cannot solve the problem after packet loss occurs Fifth, the existing error repair method uses inter-channel correlation or intra-channel correlation to predict, but does not consider the correlation between inter-channel and intra-channel prediction, and cannot make full use of lost data and Due to the complex nonlinear relationship between adjacent frames, the effect of bit error repair is not ideal, and the advantages of strong sense of space and more natural and vivid surround sound are lost.

Images