SLASH: Self-Supervised Speech Pitch Estimation Leveraging DSP-derived Absolute Pitch

TL;DR

SLASH is a novel self-supervised speech pitch estimation method that integrates DSP-derived absolute pitch and a differentiable spectrogram to improve accuracy over existing SSL and DSP approaches.

Contribution

It introduces a new approach combining self-supervised learning with DSP-derived absolute pitch and a differentiable spectrogram for enhanced speech pitch estimation.

Findings

Outperforms baseline DSP and SSL methods in pitch estimation accuracy.

Effectively predicts aperiodic components in speech signals.

Demonstrates improved robustness and applicability in speech processing.

Abstract

We present SLASH, a pitch estimation method of speech signals based on self-supervised learning (SSL). To enhance the performance of conventional SSL-based approaches that primarily depend on the relative pitch difference derived from pitch shifting, our method incorporates absolute pitch values by 1) introducing a prior pitch distribution derived from digital signal processing (DSP), and 2) optimizing absolute pitch through gradient descent with a loss between the target and differentiable DSP-derived spectrograms. To stabilize the optimization, a novel spectrogram generation method is used that skips complicated waveform generation. In addition, the aperiodic components in speech are accurately predicted through differentiable DSP, enhancing the method's applicability to speech signal processing. Experimental results showed that the proposed method outperformed both baseline DSP and SSL-based pitch estimation methods, attributed to the effective integration of SSL and DSP.