Continual Speaker Identity Unlearning with Minimal Interference

Anonymous Authors
Affiliations withheld for double-blind review
Under Submission

Abstract

Machine unlearning removes designated concepts or memorized data from pre-trained models. Recent work has made strong progress on speaker identity unlearning in zero-shot text-to-speech (ZS-TTS), but quietly assumes all unlearning requests arrive at once — an unrealistic assumption, since privacy-motivated removals arrive sequentially over time. We show this assumption breaks state-of-the-art methods: unlearning each new speaker fully revives previously unlearned speakers, reintroducing the very privacy risk unlearning was meant to eliminate. We present CORTIS (Cumulative ORThogonal Identity Suppression), the first framework for continual speaker identity unlearning in ZS-TTS that requires no access to previously unlearned speaker data. CORTIS combines Fisher-information-based parameter masking with orthogonal projection against subspaces spanned by prior unlearning updates. On VoiceBox, CORTIS unlearns each requested speaker while keeping previously unlearned speakers forgotten across long request sequences, substantially outperforming sequential application of prior methods.

Audio Demonstrations

Audio samples comparing CORTIS against baselines on forgotten and remained speakers across sequential unlearning requests.

Continual Unlearning: Forget Speakers

Each step below shows the model state after that unlearning request. Every speaker unlearned up to that point is listed, so you can verify that CORTIS maintains suppression of earlier speakers as new requests arrive. The row highlighted in green is the speaker being unlearned at that step; rows above it are prior forget speakers.

Our Methods on Remain Speakers

While our methods effectively prevent synthesis of Forget Speakers voices, it succeeds to retain the Zero-Shot performance for all other Remain Speakers. Here, the Remain Speakers are unseen voices from LibriSpeech tested in Zero-Shot setting.

BibTeX

@inproceedings{
          cortis2026,
          title={Continual Speaker Identity Unlearning with Minimal Interference},
          author={Anonymous Authors},
          booktitle={Under Submission},
          year={2026}
        }