Exploring speech style spaces with language models: Emotional TTS without emotion labels
This addresses the challenge of learning emotional prosody implicitly for TTS systems, offering a label-free approach that could benefit applications in voice synthesis, though it appears incremental in leveraging existing language models and style tokens.
The paper tackled the problem of emotional text-to-speech (E-TTS) without relying on human-annotated emotion labels, which are often inaccurate and hard to obtain, by proposing TEMOTTS, a two-stage framework that transfers knowledge from linguistic to emotional style spaces, resulting in improvements in emotional accuracy and naturalness.
Many frameworks for emotional text-to-speech (E-TTS) rely on human-annotated emotion labels that are often inaccurate and difficult to obtain. Learning emotional prosody implicitly presents a tough challenge due to the subjective nature of emotions. In this study, we propose a novel approach that leverages text awareness to acquire emotional styles without the need for explicit emotion labels or text prompts. We present TEMOTTS, a two-stage framework for E-TTS that is trained without emotion labels and is capable of inference without auxiliary inputs. Our proposed method performs knowledge transfer between the linguistic space learned by BERT and the emotional style space constructed by global style tokens. Our experimental results demonstrate the effectiveness of our proposed framework, showcasing improvements in emotional accuracy and naturalness. This is one of the first studies to leverage the emotional correlation between spoken content and expressive delivery for emotional TTS.