Learning-free L2-Accented Speech Generation using Phonological Rules

TL;DR

This paper introduces a learning-free accented TTS system that uses phonological rules to modify phoneme sequences, enabling accent transfer without requiring accented training data and allowing explicit control over phoneme-level accent features.

Contribution

It presents a novel phonological rule-based approach for accented speech synthesis that operates without training on accented datasets, offering fine-grained control and effective accent transfer.

Findings

Effective accent shift demonstrated in experiments

Maintains speech quality during accent transfer

Enables explicit phoneme-level accent manipulation

Abstract

Accent plays a crucial role in speaker identity and inclusivity in speech technologies. Existing accented text-to-speech (TTS) systems either require large-scale accented datasets or lack fine-grained phoneme-level controllability. We propose a accented TTS framework that combines phonological rules with a multilingual TTS model. The rules are applied to phoneme sequences to transform accent at the phoneme level while preserving intelligibility. The method requires no accented training data and enables explicit phoneme-level accent manipulation. We design rule sets for Spanish- and Indian-accented English, modeling systematic differences in consonants, vowels, and syllable structure arising from phonotactic constraints. We analyze the trade-off between phoneme-level duration alignment and accent as realized in speech timing. Experimental results demonstrate effective accent shift while maintaining speech quality.