Multi-Representation Knowledge Distillation For Audio Classification

Abstract

As an important component of multimedia analysis tasks, audio classification aims to discriminate between different audio signal types and has received intensive attention due to its wide applications. Generally speaking, the raw signal can be transformed into various representations (such as Short Time Fourier Transform and Mel Frequency Cepstral Coefficients), and information implied in different representations can be complementary. Ensembling the models trained on different representations can greatly boost the classification performance, however, making inference using a large number of models is cumbersome and computationally expensive. In this paper, we propose a novel end-to-end collaborative learning framework for the audio classification task. The framework takes multiple representations as the input to train the models in parallel. The complementary information provided by different representations is shared by knowledge distillation. Consequently, the performance of each model can be significantly promoted without increasing the computational overhead in the inference stage. Extensive experimental results demonstrate that the proposed approach can improve the classification performance and achieve state-of-the-art results on both acoustic scene classification tasks and general audio tagging tasks.

MeWEHV: Mel and Wave Embeddings for Human Voice Tasks

“…Another model based on the combination of representations is the model proposed by Gao et al (2020), in which they combined three types of audio representations, which fed two models, one trained for acoustic scene classification and the other for general audio tagging. They use the DCASE 2018 Challenge dataset 2 , achieving a mAP@3 of 93.3% in the acoustic scene classification task and an accuracy of 72.48% in the acoustic scene classification task, outperforming the results of other state-of-the-art methods based on a single representation.…”