Design And Implementation Of Word And Speech Libraries And NHMM Algorithm In Chinese Speech-to-Text Conversion

In the field of speech recognition, Chinese language conversion is a hot topic. Hidden Markov Model (HMM) is a commonly used method of language conversion. Because it is a good description of the speech signal of stability and variability, so in recent years, much attention of scholars home and abroad. But its recognition performance is not satisfactory. Conversion of sound in the language thesaurus library design patterns, there are many different design patterns. The different models have different rates and the proportion of space efficiency. How to find an efficient sound design pattern library thesaurus, becomes particularly important. These two aspects of this thesis will propose a new improved algorithms and design patterns to improve the conversion efficiency of language conversion.Language conversion process in the sample quantization algorithm is a research study has been difficult. And its environmental requirements for voice is also very complex. This paper intends to change the past, the traditional HMM methods, and introduced an improved recognition algorithms NHMM, to further improve the conversion efficiency of the conversion language. The traditional HMM algorithms, although widely used, but there are also some of its own defects. For example, because it is only a theory of probability and statistical algorithms, and is a discrete probability and statistics algorithms. It can not well describe the speech signal of the time dependencies. And the quantization error analysis does not focus on consideration. This will reduce the language to some extent, the recognition rate of conversion. Therefore, this paper proposes a method of weighting function can be introduced-NHMM algorithm. NHMM algorithm in this paper is based on the HMM algorithm proposed an improved algorithm. It is to further reduce the speech signal in the quantization process large errors occur. In order to improve the recognition rate of speech signal in the NHMM algorithm, added a new variable-the quantization error E, as a weighted value, added to the HMM algorithm for parameter sequence. HMM algorithm and make it as a parameter in operation. This will quantify the error as a factor in the improved algorithm in HMM recognition rate, compared to the traditional HMM algorithms have greatly improved. In the field of speech recognition, Chinese language conversion is a hot topic. Hidden Markov Model (HMM) is a commonly used method of language conversion. Because it is a good description of the speech signal of stability and variability, so in recent years, much attention of scholars home and abroad. But its recognition performance is not satisfactory. Conversion of sound in the language thesaurus library design patterns, there are many different design patterns. The different models have different rates and the proportion of space efficiency. How to find an efficient sound design pattern library thesaurus, becomes particularly important. These two aspects of this thesis will propose a new improved algorithms and design patterns to improve the conversion efficiency of language conversion.In the design of speech Library and word Library, we have access to a large number in the Chinese language on the process of converting the relevant literature, we found lot of research is focused on the conversion of audio language library design. And library files are sought to the maximum sound coverage. But the detailed design of the audio library database will inevitably lead to the rapid increase of body size. This hardware configuration of the terminal made a very big challenge. If you take a lot of system resources, will inevitably lead to the feasibility of decline. Starting point of this research is to try to streamline the language conversion and audio library. This will inevitably lead to language conversion, on the accuracy of people's spoken language input was ignored for consideration. Sampling from the rough design ideas of fuzzy sampling accuracy, and further improve the matching language thesaurus conversion efficiency. The design of the body through three libraries and Disadvantages of the structure, and ultimately determine the most efficient body design pattern library. In the final method, the thesaurus structure, three-stage format, namely, word form, double-word list and thesaurus. In this design pattern, the language conversion without compromising the accuracy of the implementation of the premise of the roughness of the voice recording, storage capacity, while reducing sound and detailed structure of word library. The proposed conversion of the thesaurus based on streamlined design language with a strong novelty and practicality.