Chinese and English mixed offline voice keyword recognition method in specific scene and system implementation thereof

Abstract

The embodiment of the invention discloses a Chinese and English mixed offline voice keyword recognition method and system. A specific embodiment of the method comprises the following steps: acquiring a voice digital signal, and performing voice activity detection on the voice digital signal to obtain a voice segment to be recognized; defining a self-adaptive keyword matching window, and segmenting a voice segment to be recognized; performing feature extraction on the voice segment to obtain a Mel-frequency cepstrum coefficient embedded feature vector; analyzing the user-defined keyword list, and obtaining a Chinese decoding network space and an English decoding network space in combination with a pre-trained phoneme whitening model; sequentially inputting the Mel-frequency cepstrum coefficient embedded feature vectors into the decoding network space to obtain an identification result; and performing post-processing on the identification result to generate a target identification result. According to the embodiment, the calculation amount is low, offline recognition can be achieved, the recognition accuracy is high, the response speed is high, Chinese and English mixed recognition is supported, and the keyword list can be flexibly replaced so that the method can adapt to application in different scenes.

Description

technical field [0001] The embodiments of the present disclosure relate to the technical field of speech recognition, and in particular, to a method and a system for offline speech keyword recognition in Chinese and English. Background technique [0002] Speech keyword recognition technology is a research branch in the field of automatic speech recognition. Automatic speech recognition technology requires complete decoding and conversion of the collected continuous speech streams, and requires higher hardware computing capabilities. Online recognition of data uploading to cloud computing is often used. When the network is offline, the recognition effect drops sharply, and online recognition also has the risk of privacy leakage when data is uploaded to the cloud. Speech keyword recognition only pays attention to the keyword part, which is less dependent on hardware. It can complete the decoding calculation locally, so as to realize offline recognition. It has broad applicatio...

AI Technical Summary

Problems solved by technology

Voice keyword recognition based on embedding learning is to train a neural network feature extractor (for example: LSTM feature extractor), convert the voice to be recognized and the keyword template voice into feature vectors of the same length through the feature extractor, and then calculate Vector distance, this implementation method has a high recognition accuracy rate when identifying a single keyword, so it is widely used in the field of smart device wake-up, but as the number of keywords to be identified increases, the recognition accuracy rate will drop sharply. Although it is only necessary to collect a small amount of keyword template corpus to replace the keywords to be recognized, this method also brings about the problem of poor recognition of non-specific people.

Images