Experiment Research On Blind Speech Separation For Intelligent Vehicle Applications

In recent years,with the development of Internet of Things(Io T)and artificial intelligence(AI)technologies,human-computer interaction based on speech recognition has received more and more attention.However,when intelligent devices acquiring voice commands,the influence of interference signals and environmental noise is considerable,which in turn affects the efficiency of command recognition.Meanwhile,most of the current speech recognition systems can only recognize a single voice command at the same time.When multiple voice commands are input simultaneously,it cannot respond correctly.In this paper,we explored the separation of mixed speech signals in actual environments.At first,the theoretical basis of the microphone array and blind source separation technology is analyzed,and the feasibility of related technologies in this design is proved.Then,the hardware system for audio acquisition and transmission was designed and realized.A uniform linear microphone array with 8 elements and 5cm spacing acts as the signal acquisition front-end.Based on FPGA,we drive each array element,sample and decode raw data synchronously.The Wi Fi module is used to transparently transmit the data stream to the host computer for further processing.The test result shows that the hardware system can realize the required functions and has certain scalability.Subsequently,the independent vector analysis(IVA)algorithm for the convolutional mixture model is introduced,which transforms the signal into the complex frequency domain firstly.Then,we treat all frequency components as a multidimensional vector for overall iteration and separation,which overcomes the inherent problem of independent component analysis,permutation uncertainty.At last,the signal is inversely transformed back to the time domain to obtain the output result.Finally,we collect mixed speech signals in the vehicle environment and perform IVA algorithm.The experimental results and error analysis show that: in practical environment,the system can separate 5 signals with good signal quality.When there are less target sources to be separated,the quality of the separated signal is better.This article has realized the collection and separation of audio signals in actual environment,which has certain practical significance.