Lightweight speech emotion recognition method and system based on multi-scale convolution

The invention discloses a lightweight speech emotion recognition method and system based on multi-scale convolution, and relates to the technical field of speech emotion recognition. The method comprises the following steps: acquiring and preprocessing to-be-tested voice data, and extracting a Mel-frequency cepstrum coefficient of the to-be-tested voice data; processing the Mel-frequency cepstrum coefficient by using a speech emotion recognition network to obtain a speech emotion recognition classification result, in the speech emotion recognition network, a lightweight multi-scale feature extraction module being used for adaptively learning features of different scales under element-level granularity and performing multi-scale feature extraction to obtain a speech emotion recognition classification result; the multi-scale cepstrum and time spectrum attention module is used for sequentially optimizing the key cepstrum component of the Mel-frequency cepstrum coefficient and the key time spectrum position of th