【Abstract】 Hidden Markov Model (HMM) is widely used in modern speech recognition systems. The HMM model of continuous density distribution often assumes the covariance matrix to be a diagonal matrix for the sake of computational complexity and model simplification, but this will undoubtedly reduce the recognition performance. Under the premise of ensuring a high recognition rate of speech signals, this paper adopts a new model based on SPM (segmented probability model). This model does not need to be vector quantized. It also has the characteristics of continuous density distribution, but does not have such a large amount of computation. It is the continuous distance density segmented probability model (CDD-SPM). This paper mainly studies the speech recognition system based on AD1836 and ADSP21161 and its software design. The full paper analyzes the linear prediction cepstral coefficient (LPCC), the hidden Markov model (HMM) and the segmented probability model of continuous distance density (CDD-SPM) based on it, and introduces in detail the processing unit based on ADSP21161 EZ-KIT-Lite development board in the speech recognition system and the software design of the speech recognition system, such as the starting point judgment, speech sample feature parameter extraction, speech signal training and recognition system. According to the final test data and system analysis, the technical indicators of this system fully meet the initial design requirements of this design system. The main work completed in this paper: (1) Analyzed the shortcomings of the training and recognition distance model based on the hidden Markov model (HMM) in terms of operation speed, selected the CDD-SPM algorithm idea that saves clock cycles, and analyzed the characteristics of the speech signal...
You Might Like
Recommended ContentMore
Open source project More
Popular Components
Searched by Users
Just Take a LookMore
Trending Downloads
Trending ArticlesMore