Voice content processing method and device, equipment and readable storage medium

Abstract

The invention discloses a voice content processing method and device, equipment and a readable storage medium, and relates to the field of machine learning. The method comprises the following steps: acquiring voice content; performing feature extraction on the voice content to obtain an audio feature of a first data format; dynamically quantizing the audio feature to obtain a quantized feature of a second data format, wherein the data bits of the first data format are greater than the data bits of the second data format; and inputting the quantitative feature into a voice processing model, and outputting to obtain a content processing result. Model parameters in the voice processing model are quantized from the first data format to the second data format, and audio feature is quantized to be in the second data format before audio feature processing; and since the data bits of the first data format are greater than the data bits of the second data format, the overall data occupation amount of the voice processing model is reduced, and the external memory (such as flash) and memory occupation conditions of the voice processing model in the mobile device are reduced.

Description

technical field [0001] The embodiments of the present application relate to the field of machine learning, and in particular to a method, device, device, and readable storage medium for processing voice content. Background technique [0002] With the rapid development of the field of machine learning, there is an increasing demand for using offline neural networks on mobile devices. For example, in offline scenarios, using the neural network models on mobile The development of network algorithms has led to an increasing demand for computing and memory by neural networks, so that the computing power and memory space of mobile devices cannot bear it. [0003] In related technologies, taking the speech recognition scene as an example, the acoustic model is usually trained by using the neural network TensorFlow and Pytorch framework, and embedded in the speech framework Kaldi, so as to realize the quantization of the acoustic model. [0004] However, the implementation cost of ...

AI Technical Summary

Problems solved by technology

[0004] However, the implementation cost of the above method is relatively high. The TensorFlow and Pytorch frameworks lack speech decoder-related technologies. Even if they are integrated into the speech framework Kaldi, due to the integration of the two frameworks, the memory usage will inevitably increase, and the transformation and maintenance costs will also be large.

Images