A Real-Time Speech Recognition System Based On The Implementation Of FPGA

Speech is one of the mainly way between human interaction and communication. With the advance of speech technology, a variety of speech products, such as mobile phone, PDAs and so on, which have been gradually known. With the popularity of wireless communication and network, speech products would have been severed as the primary interface for communication between the new generation of intelligent products and human. The products of speech recognition are required a higher real-time and integration, the design methodology which combined the design of digital integrated circuit with the design of embedded system were used for the implement of embedded real-time speech recognition system, and this research had important significance.First, this paper introduced the mainly algorithm of speech recognition theorem, which included MFCC algorithm、Vector Quantification algorithm、Baum-Welch algorithm and viterbi search algorithm and so on.The ten digit system of hidden Markov model (HMM) was established among0-9by calling Matlab based on HTK, which parameters of MFCC were set for13-dimension. The speech data of speaker-dependent and speaker-independent were trained and recognized, speaker-dependent speech recognition was85.33%, and the result of speaker-independent speech recognition was75.12%.The visual system of speech recognition was established by calling HTK software based on VC, which was implemented to the extract of speech feature and display of recognition result. The training databases were downloaded to the SD card of Field-Programmable Gate Array (FPGA), which was used as comparison of library of real-time speech recognition.The designed system was used a custom SOPC system, which was integrated the C program module into NIOS-Ⅱ system. The self-defined IP cores and multiple parallel input output (PIO) interface were set in order to the interconnection of hardware modules and embedded systems, which made full of the advantage of SOPC system parallel computation in the digital signal process.The SOPC technology of FPGA was used to establish embedded real-time speech recognition system, which the whole system was integrated into a single chip in the chip of DE2-70EP2C70F896C6. A real time speech recognition engine was implemented that takes as an input the time domain signal from a microphone and performs the frequency domain feature extraction on the sample to identify the word being spoken.The experimental results show that speech recognition result was the same as software recognition result, the execution time of FPGA was nearly26times faster than the execution time of software, this system of speech recognition meet the requirements of real-time speech recognition system.