Multi Channel Audio Signal Recovery Based On Low Rank Matrix Completion

As a common audio signal carrier,multi-channel is widely used in music,film production and home theater construction.Through the cooperation of multiple channels,people can experience a deeper sense of level and three-dimensional surround in the music,and play a good role in the scene reproduction of live music shows such as concerts and livehouses.In the process of film production,the use of multi-channel audio technology can make the audience experience the scene through different angles and levels of sound placement in building a home theater,multi-channel audio technology plays a fundamental role in the construction of stereo surround sound.Multichannel audio signal is collected by placing radio equipment in different positions and matching audio acquisition technology.Due to the large amount of data and the strong correlation between the channels,part of the data may be lost or damaged in the process of acquisition,storage,transmission and use,so that part or all of the channels of the multi-channel audio signal can not work normally,thus affecting the auditory feeling of the audience.If the damaged multi-channel audio signal is used in speech recognition,the incomplete data will reduce the recognition rate.In order to solve this problem,this paper uses a singular value shrinkage algorithm based on low rank matrix completion to complete and recover multi-channel audio signal.Firstly,the collected multi-channel audio signal is cleaned and preprocessed,and the effect of data loss is achieved through random sampling;then the multi-channel audio signal with data loss is represented by low rank,and the Lagrange operator is obtained by solving the kernel norm minimization problem,and the convex optimization technique;finally,the operator is updated continuously in the process of updating The new singular value is obtained by singular value decomposition(SVD).Compared with the singular value threshold,the singular value which is larger than the singular value threshold is removed and updated continuously.Finally,the singular value is reduced gradually,and the recovered audio signal is obtained.Singular value decomposition(SVD)can not only reduce the dimension of data processing effectively,but also is a decomposition method which can be applied to any matrix and has more extensive applicability.In the recovery experiments with different loss rates,the comparison with other audio signal recovery methods shows that:1)With the same number of iterations,the SVD algorithm based on low rank matrix completion can recover the multi-channel audio signal with missing data with higher accuracy;2)The update iteration mode of SVS algorithm is more suitable for the data volume of multi-channel audio signal.Therefore,the audio signal recovered by the algorithm in this paper is closer to the original complete audio.In the evaluation system of multi-channel audio recovery results,this paper uses the combination of objective evaluation and subjective evaluation.Objective evaluation is obtained by comparing the signal-to-noise ratio of audio signal after data loss with that of audio signal after recovery.Using SNR evaluation can not only directly reflect the success of audio signal recovery from the data,but also reflect the recovery degree and effect of the data through undifferentiated comparison;subjective evaluation uses a multi incentive hidden reference mushra test method,through the audience's real experience of the multichannel audio signal after the loss of data and recovery,and then make a percentage analysis The evaluation results are obtained by scoring.The advantage of mushra is that it can take the audience as the main body and evaluate the recovery results hierarchically.To sum up,this paper carries out the research and evaluation based on low rank matrix completion method for multi-channel audio signal recovery.The research results can not only expand the research content of audio signal recovery technology,but also lay the foundation for the application of low rank matrix completion in the field of audio signal research,which has certain practical value.