Method, device, equipment and medium for speech recognition

Abstract

Embodiments of the present invention provide a voice recognition method, device, equipment and medium. The method includes: converting the acquired audio data into a corresponding spectrogram; judging whether the frame number of the spectrogram is a preset frame number; if the frame number of the spectrogram is not the preset frame number , then the spectrogram is zero-filled, so that the number of frames of the spectrogram to be recognized obtained after zero-padding is the preset frame number; the spectrogram to be recognized is input to the multi-task convolutional neural network network acoustics model. It realizes the direct input of spectrogram to the acoustic model, and then recognizes the text of the audio data. Compared with the information loss in the frequency domain caused by the calculation of MFCC features in the prior art, the present invention reduces the loss of input features, increases the recognition of audio data, and is more conducive to the extraction of feature information by acoustic models.

Description

technical field [0001] The present invention relates to speech recognition technology and deep learning technology, in particular to a speech recognition method, device, equipment and medium. Background technique [0002] With the popularity of intelligent products, speech recognition technology as a human-computer interaction is becoming more and more important. [0003] In speech recognition, most of the traditional speech features are used for speech recognition at present. Among them, traditional speech features include: MFCC features, FBANK features and other artificially designed features, which cause information loss in the frequency domain, especially in the high-frequency region, resulting in low accuracy of speech recognition. At the same time, the traditional single-task network model can easily overfit on the training data, resulting in a decline in the recognition rate on the test set. Contents of the invention [0004] In view of the above problems, embodim...

AI Technical Summary

Problems solved by technology

Among them, traditional speech features include: MFCC features, FBANK features and other artificially designed features, which cause information loss in the frequency domain, especially in the high-frequency region, resulting in low accuracy of speech recognition

At the same time, the traditional single-task network model can easily overfit on the training data, resulting in a decline in the recognition rate on the test set.

Images