Method and Network for Time-Frequency Channel Attention Weight Calculation and Vectorization

Abstract

This application relates to the method and network of time-frequency channel attention weight calculation and vectorization, including: using spherical embedding normalization to preprocess the logarithmic Fourier magnitude spectrum; segmenting the logarithmic Fourier magnitude spectrum preprocessing Process data into short-term spectral segments to generate multi-channel tensors; use the attention mechanism to calculate the attention weight of each channel in the time direction and frequency direction; obtain time-frequency attention by matrix multiplication of the attention weights in the time direction and frequency direction Weight coefficient tensor; use the attention mechanism to calculate the attention weight tensor of each channel in the time-frequency attention weight coefficient tensor; use the attention weight tensor of each channel to obtain vectorization in the time and frequency directions through one-dimensional convolution Results; concatenate the vectorized results of the time direction and frequency direction of at least one channel, and use one-dimensional convolution to obtain the channel vectorized results; calculate the average value of the channel vectorized results of the short-term spectrum segment and use it as the entire long-term The results of the time-to-speech correspondence.

Description

technical field [0001] This application relates to the field of artificial intelligence, in particular to a method and network for time-frequency channel attention weight calculation and vectorization. Background technique [0002] The purpose of automatic depression detection is to explore the changes in the speech of healthy individuals and depressed patients, and propose corresponding models and methods to establish the ability of the machine to capture clues of depression, enhance its diagnostic ability, and improve diagnostic efficiency. Automatic depression detection is an important research direction in the field of human-computer interaction and artificial intelligence, involving many fields such as intelligence science, mathematics, psychology, and physiological science. [0003] Physiological and psychological studies have shown that different frequency segments and time segments of speech have different effects on depression detection. In other words, it is neces...

AI Technical Summary

Problems solved by technology

Many current methods treat all frequencies and time periods of raw speech indiscriminately to predict levels and fail to target discriminatively relevant frequency bands and time periods to capture differential cues between healthy individuals and depressed patients

Images