Toward end-to-end interpretable convolutional neural networks for waveform signals

TL;DR

This paper proposes a new CNN framework for end-to-end audio analysis that improves efficiency and interpretability, outperforming traditional features in speech emotion recognition and capturing complex waveform patterns.

Contribution

Introduces a novel CNN framework for raw waveform data that enhances interpretability and efficiency, outperforming traditional spectral features in speech emotion recognition.

Findings

Outperforms Mel spectrogram features by up to 7% in speech emotion recognition

Demonstrates effective handling of long waveform patterns in heart sound data

Provides a lightweight, interpretable model for raw waveform analysis

Abstract

This paper introduces a novel convolutional neural networks (CNN) framework tailored for end-to-end audio deep learning models, presenting advancements in efficiency and explainability. By benchmarking experiments on three standard speech emotion recognition datasets with five-fold cross-validation, our framework outperforms Mel spectrogram features by up to seven percent. It can potentially replace the Mel-Frequency Cepstral Coefficients (MFCC) while remaining lightweight. Furthermore, we demonstrate the efficiency and interpretability of the front-end layer using the PhysioNet Heart Sound Database, illustrating its ability to handle and capture intricate long waveform patterns. Our contributions offer a portable solution for building efficient and interpretable models for raw waveform data.