基于凌陽單片機(jī)的語音識別系統(tǒng)-在線瀏覽
2024-09-20 17:14本頁面
【正文】 d on Lingyang Single ChipAs a munication technology between manmachine interactive technology, voice recognition is widely used. This paper introduces the design of voice recognition for several orders, namely isolated words, and for special person. In the practical process, the design regards short Chinese orders as objects of recognition, embodying Chinese voice traits Moreover, it sresses simple and practical characteristics in the software and hardware design, simplifies the system structure and strengthens the control capacity. This method overes the defects of the similar system ertablished on DSP chips, such as plex design, high price and inconvenient control.1 INTRODUCTIONAccording to the different of actual needs and applications, Speech Recognition can be divided into isolated word recognition and continuous speech recognition, specific recognition and nonspecific recognition. The main indicators of the Speech Recognition pursuit is the high recognition rate, realtime and large vocabulary. But for a voice recognition system, we should also consider software and hardware design simple, cheap, flexible extemal control, humanputer interaction and other convenient features. Now, the main of the applied to speech recognition chips is DSP (digital signal processor chip ), for example TI′TMS320 series. However, if DSP chips be used for a small voice recognition system ,its inadequacies is very obvious :(1) pins , high prices, the use of red tape。(2) weak control, it needs bine with SCM or FPGA ( field programmable gate arrays ) to achieve humanputer interaction 。(4) V for the pin, we must consider the level match when it connections with SCM, FPGA, Flash memory and so on 。 IF for a fixed value, usually the value of the IF is 25。 the first frame which the three of the 10 frames in a row is more than IZCT and 2 frames over the IIL —the situation posed by Qingyin and Zhuoyin. The end discrimination: the first frame which consecutive 5 frames of the En and ZCRn are less than threshold —no word at the end of Qingyin, sampling not too long, but we must prevent a misjudgment of the word among Qingyin. Feature Extraction Compared with other voice characteristics, LPCC (Linear Prediciton Cepstrum) recurrence formula, speed and accuracy is better, particularly suitable for the word recognition of the specific isolation of the short time. LPCC is obtained on the basis of the LPC (coefficient characteristics of Linear Prediction):c (1) = a (1)c (n) = a (k) c ( n k ) + a ( n )In the above two formulas: c (n) （ n = 1,2, … ,p）,coefficient for the LPCC。 12 bands will take to the vast majority of the voice signal channel model enough approximating,a (k) is the fearures of the LPC, 1＜n≤p。 create templates under learning, identify under the discrimination functions. (2) Acquisition of the voice signalThere are microphone and the signalchannel sound A/D converter of the AGC function, so it can omit many frontend processing hardware, simplify the circuit, improving stability. Conect with the microphone and antialiasing filter ( 100Hz4000Hz bandpass filter ), access the channel, pleted 108 kHz sampling signal. (3) Some expansion of data storageBecause of voice signal processing needs the larger data storage, so the extended 32KB Flash memory as a data memory. The storage space is divided into four parts: template storage area for storing the leaning process by isolating the characteristics of the word template, the storage area the size of the number of templates stored decision, it decided to identify the vocabulary。 the intermediate data storage area contains builtin 2 KB SRAM, storage middle accounts, such as background noise characteristics, the middle templates of the learning process。 then it sample 52 voice datas and enter the judgement: first, we must do the judgement of the end point to determine the scope, if there is not point, then error handling。 then prompted to “ the first voice
高考資料相關(guān)推薦
鄂ICP備17016276號-1