Research On Computer - Aided Field Investigation System Based On

China is a multi-ethnic, multilingual country, language research facing a very arduous task. The development of computer software in recent years more and more for field investigation of language, greatly improve the efficiency of investigation. However, in the computer software in the finished recording work, recording and proofreading links also remain in the stage of perception. Because the number of field investigation data recording large, heavy workload, difficult to maintain a consistent standard, resulting in the fallacy of human reason, and there is no more effective way to reduce the workload of recording. This paper studied the single word recognition based on HTK feature analysis, the single word recording files on the existing field investigation corpus preprocessing, MFCC feature extraction, HMM model and establish corresponding to each sound, speech feature extraction using HTK toolkit to realize acoustic model training, finally identify performance test. This paper collection and pretreatment on the field investigation data, with the same pronunciation in different languages, and for the MFCC dimension, HMM number and GMM number of state of the three parameters of recognition of single word rate. Experiments show that for the single word contains the dialect and the phonetic materials national language more languages, for voice data in a single independent training quantity is sufficient, the recognition rate can reach about 80%. And in the experimental material under the increase of HMM number and GMM number of state is conducive to the improvement of recognition rate, but more than a certain value, the recognition rate will decline. For tone recognition for single speech recognition rate and the training is proportional to the amount of. In the speech data is sufficient, can get ideal recognition rate.