Construction method of heterogeneous decoding network, system thereof, voice recognition method and system thereof

Abstract

The invention discloses a construction method of a heterogeneous decoding network, a system thereof, a voice recognition method and a system thereof. The construction method comprises a step of obtaining a command word decoding network and a dictation decoding network needed by service, a step of constructing a command word training data set and a dictation training data set, a step of using the command word training data set and the dictation training data set and determining the compensation score of the command word decoding network, a step of optimizing the command word decoding network and storing a command vocabulary entry, a compensation score and the acoustic model scores corresponding to the command vocabulary entry for each arc of the optimized command word decoding network, and a step of carrying out the parallel recognition of the command word decoding network and the dictation decoding network to form the heterogeneous decoding network. After the application of the construction method of the heterogeneous decoding network and the system of the embodiment of the invention, the parallel recognition of different services and the saving of system resources are completed through configuring a set of recognition engine. Through the application of the construction method and the system of the embodiment of the invention, the recognition effect and recognition efficiency can be improved, and the complexity of the system is reduced.

Description

technical field [0001] The present invention relates to the technical field of speech signal processing, in particular to a method and system for constructing a heterogeneous decoding network, and a method and system for speech recognition. Background technique [0002] Speech recognition technology is a technology that allows machines to convert voice signals into corresponding text or commands through the process of recognition and understanding. Existing speech recognition systems usually can only support command word recognition (rule-based speech recognition system) or dictation recognition (statistic-based speech recognition system) alone. Therefore, in some multi-service scenarios (such as the application on vehicle equipment, both command word recognition is required, such as "call xxx", "send text message to xxx", and dictation recognition is also required, such as the content of text messages, navigation Destinations, etc.; another example is the use of voice assi...

AI Technical Summary

Problems solved by technology

This method will not only occupy a large amount of system resources and affect system performance, but also, because the structures of the dictation decoding network and the command word decoding network are inconsistent, confidence judgment is required to select the final recognition result, which has poor reliability and low efficiency, and increases the The complexity of the entire speech recognition system

Images