Zero-Shot Text-to-Speech for Text-Based Insertion in Audio Narration

TL;DR

This paper introduces a novel one-stage, zero-shot text-to-speech editing framework that generates natural speech for audio narration without requiring target speaker training data, outperforming existing methods.

Contribution

The proposed approach eliminates the need for target speaker training data by using a context-aware, transformer-based model with zero-shot duration prediction for seamless speech editing.

Findings

Achieves high-quality speech synthesis without target speaker data

Outperforms recent zero-shot TTS engines in subjective tests

Provides accurate zero-shot duration prediction for inserted text

Abstract

Given a piece of speech and its transcript text, text-based speech editing aims to generate speech that can be seamlessly inserted into the given speech by editing the transcript. Existing methods adopt a two-stage approach: synthesize the input text using a generic text-to-speech (TTS) engine and then transform the voice to the desired voice using voice conversion (VC). A major problem of this framework is that VC is a challenging problem which usually needs a moderate amount of parallel training data to work satisfactorily. In this paper, we propose a one-stage context-aware framework to generate natural and coherent target speech without any training data of the target speaker. In particular, we manage to perform accurate zero-shot duration prediction for the inserted text. The predicted duration is used to regulate both text embedding and speech embedding. Then, based on the aligned cross-modality input, we directly generate the mel-spectrogram of the edited speech with a transformer-based decoder. Subjective listening tests show that despite the lack of training data for the speaker, our method has achieved satisfactory results. It outperforms a recent zero-shot TTS engine by a large margin.