Speech Recognition Algorithm Based On Microphone Array Model Compensation

The speech recognition technology is one of the most important signal processingtechnology, which has a wide range of applications in other fields of moderncommunication,multimedia and artificial intelligence systems.Existing speech recognitionsystems are usually designed for the controlled environment of pure speech,but theperformance of speech recognition systems suffer a sharp decline in noisy environment withthe pollution of noise.How to make the speech recognition system not affected by ambientnoise and maintain a high recognition rate has become a key issue to be resolved.Model compensation algorithm correct the characteristic parameters of the model of purespeech according to the environmental characteristics of the noise and compensate thedifference between the training environment and test environment in order to make the voicemodel after the compensation approach the actual distribution with noisy speech as much aspossible. So by using the compensated voice model, it can greatly improve the systemrecognition rate.A problem existing in model compensation is that when the noisy signal SNRof the actual test environment decreases and it makes the model parameters difficult todistinguish，resulting in a serious decline in recognition results.Model compensation is aeffective back-end processing means in the model layer,while microphone array speechenhancement is a front-end processing means of speech in the signal layer.Microphone arraywith good spatial selectivity can effectively suppress noise interference in addition to thedirection of the desired speech signal so as to achieve significant noise cancellation. Inconclusion,the speech recognition system combining the front-end microphone array in theback-end model compensation module can effectively obtain better recognition performance.This thesis combines microphone array speech enhancement algorithm(front-endprocessing) and model compensation algorithm(back-end processing) and enhance the speechsignal by using microphone array so as to reduce the noisy components of the noisy signal toachieve satisfactory identification when noisy signal SNR reduces in test conditions of modelcompensation.We enhance the speech signal with microphone array, and then extract thespeech feature for modelling, and finally calculate the recognition rate of speech recognition.Fianlly we simulate the experiment by using MATLAB and HTK tools in different SNRunder different noisy environments. The experimental results show that the new speechrecognition algorithm based on the microphone array model compensation improves thesystem recognition rate, especially in low SNR conditions.