Speech Recognition And VUI System Design Under The High Noise Background

With the rapid development of human-machine interaction technology, VUI system has gradually become a research hotspot at home and abroad. In the VUI system, change the traditional keyboard input pattern instead of human voice, which make human-computer interaction more convenient and humanization.However, due to the mismatch between the training and recognition environment caused by complex environmental noise, the VUI’s speech recognition rate will dramatically degrade when put to actual use. Thus, this article combines the Empirical Mode Decomposition, Hilbert-Huang Transform with the dual-microphone noise interference cancellation technology to improve the recognition rate under noisy environment. By firstly applying dual-microphone noise interference cancellation technology to filter part of the background noise and improve the signal-to-noise rate of the voice commands collected. Then use the empirical mode decomposition and Teager energy operator for precisely voice endpoint detection. As a result, the VUI system recognition rate under high noise environment is improved, which provides reliable human-machine interaction for aircraft maintenance. The main research contents are as follows:Firstly, this paper analyzes the research situation and development trend of VUI system at home and abroad, the current demand for civil aircraft maintenance is analyzed and the practical problems that exist is presented.Secondly, previous speech endpoint detection algorithms generally analyze short-time energy and short-time average zero crossing rate. Energy calculation method is not reasonable and recognition effect is poor under the condition of low SNR when use these characteristic parameters. In this paper, we study the speech endpoint detection technology based on EMD and Teager Energy Operator. This algorithm combines the advantage of EMD and Teager Energy Operator in dealing with nonlinear, non-stationary signal; use EMD to realize preliminary de-noising, Teager energy is then used as a substitute for short-time energy for endpoint detection.Then, traditionally we use single channel microphone to collect the voice mixed in noise and a series of processing method on the spectrum for filtering, such as wavelet transform, spectral subtraction and so on. Considering that the background noise in the aircraft maintenance scene has larger amplitude and more wide frequency domain, the thesis gives an introduction to dual-microphone noise interference cancellation technique; one microphone for collecting the voice with noise, and another microphone for background noise. Adopting RLS adaptive algorithm to cancel the two ways of signal and keep the effective voice, eventually realize the aim of de-noising.Finally, this paper elaborates the detailed implementation process of full module of VUI system and communication between them. The design adopts the client- server(C/S) structure, which effectively use the client and server load. By adaptively train the HMM model for 10 times, the VUI system automatically adjust speech template and noise threshold, the test experiment is given in this paper. Analysis shows that the VUI system has stronger anti-noise performance, and gains nearly 5% increase on recognition rate.