Eta-WavLM: Efficient Speaker Identity Removal in Self-Supervised Speech Representations Using a Simple Linear Equation

Giuseppe Ruggiero, Matteo Testa, Jurgen Van de Walle, Luigi Di Caro

Abstract: Self-supervised learning (SSL) has reduced the reliance on expensive labeling in speech technologies by learning meaningful representations from unannotated data. Since most SSL-based downstream tasks prioritize content information in speech, ideal representations should disentangle content from unwanted variations like speaker characteristics in the SSL representations. However, removing speaker information often degrades other speech components, and existing methods either fail to fully disentangle speaker identity or require resource-intensive models. In this paper, we propose a novel disentanglement method that linearly decomposes SSL representations into speaker-specific and speaker-independent components, effectively generating speaker disentangled representations. Comprehensive experiments show that our approach achieves speaker independence and as such, when applied to content-driven tasks such as voice conversion, our representations yield significant improvements over state-of-the-art methods.

Voice Conversion Demo - LJSpeech

In this section, we present speech samples generated by the voice conversion system (Section 3.2), using the cleaner LJSpeech dataset as the target speaker and the LibriSpeech test-clean dataset as the source speech.
We compare Proposed Eta-WavLM against five baselines: WavLM, Perturbation, Soft, Utterance Std, and Vector Quantization.

LJSpeech Target (F):

Scenario	Source Speech	WavLM	Perturbation	Soft	Utterance Std	Vector Quantization	Proposed Eta-WavLM
M -> F
M -> F
F -> F
F -> F

Voice Conversion Demo - Elliot Miller

In this section, we present speech samples generated by the voice conversion system (Section 3.2), using the more challenging Elliot Miller dataset as the target speaker and the LibriSpeech test-clean dataset as the source speech.
We compare Proposed Eta-WavLM against five baselines: WavLM, Perturbation, Soft, Utterance-Std, and Vector-Quantization.

Elliot Miller Target (M):

Scenario	Source Speech	WavLM	Perturbation	Soft	Utterance Std	Vector Quantization	Proposed Eta-WavLM
M -> M
M -> M
F -> M
F -> M