SonoEdit: Null-Space Constrained Knowledge Editing for Pronunciation Correction in LLM-Based TTS

TL;DR

SonoEdit is a novel model editing method that precisely corrects pronunciation errors in pre-trained TTS models without retraining, using null-space constrained updates to preserve overall speech quality.

Contribution

The paper introduces Null-Space Pronunciation Editing, a single-shot, mathematically grounded approach for targeted pronunciation correction in TTS models without extensive data or retraining.

Findings

Effective correction of pronunciation errors in low-resource proper nouns.

Preserves the overall speech quality and model behavior.

Requires only a single parameter update, reducing computational costs.

Abstract

Neural text-to-speech (TTS) systems systematically mispronounce low-resource proper nouns, particularly non-English names, brands, and geographic locations, due to their underrepresentation in predominantly English training corpora. Existing solutions typically rely on expensive multilingual data collection, supervised finetuning, or manual phonetic annotation, which limits the deployment of TTS systems in linguistically diverse settings. We introduce SonoEdit, a model editing technique that surgically corrects pronunciation errors in pre-trained TTS models without retraining. Instead of costly finetuning or explicit phoneme injection, we propose a parsimonious alternative based on Null-Space Pronunciation Editing, which performs a single-shot parameter update to modify the pronunciation of specific words while provably preserving all other model behavior. We first adapt Acoustic Causal Tracing to identify the Transformer layers responsible for text-to-pronunciation mapping. We then apply Null-Space Constrained Editing to compute a closed-form weight update that corrects the target pronunciation while remaining mathematically orthogonal to the subspace governing general speech generation. This constrained update steers the model's acoustic output toward a desired pronunciation exemplar while guaranteeing zero first-order change on a preserved speech corpus.