Study Of Speaker-independent Speech Recognition Base On Hmm/ANN Model

The purpose of the speech recognition is to make the computer understand thesemantic content of the human language.In recent years, speaker-independent isolated-word recognition (SIIWR)technology got a quick development, and it has been widely applied in interactionbetween human and machine, intelligent phone, remote control of home electricequipment. There are many key technologies used in speech recognition, mainlyincluding dynamic time warping (DTW), Hidden Markov Model (HMM) and ArtificialNeural Model (ANN), which will be discussed in this paper. This paper combinedHMM model with ANN model and proposed a new model called HMM/ANN, using thestrong ability of HMM/ANN in time domain modeling and pattern recognition toimprove the speech recognition rate and performance. Compared with the traditionalspeech recognition models, the experimental results showed the recognitionperformance and system robustness has certain improvement by the MATLABsimulation tool.Speech recognition technology at home and abroad was studied systematically;especially the research of SIIWR was done. Tasks of this paper are arranged by thefollowing:1. The Principle and algorithm of HMM was studied, and the defect was pointedout. Then the optimization method about HMM model was given based on the design inthis paper.2. The ANN model applied in speech recognition was studied, and a new neuralnetwork model for speech recognition was established.3. The improvement program was proposed based the existing problem in currentSpeech recognition field, and the non-specific isolated-word speech recognition systemwas design.