Text-to-speech models can steer rising or falling pitch intonation
Steering intonational tunes in text-to-speech synthesis
SoundComputation and LanguageMachine Learning
Summary
Text-to-speech systems create spoken words from written text, but controlling how the speech sounds—like making a question rise in pitch or a statement fall—can be tricky. The authors studied parts of a popular speech model to find which piece controls these pitch patterns, called intonation. They discovered that a specific hidden state in the main part of the model carries the tune, and by adjusting this, they could shift the speech to sound more like a question or a statement. While this steering changes the overall melody, it isn't perfect and sometimes slightly reduces naturalness.
What this means in practice
- •For voice assistant developers: Adjust intonation in speech output to better distinguish questions and statements for clearer user interaction.
- •For audiobook production teams: Modify synthesized speech prosody to make narrations sound more natural by controlling rising and falling pitch patterns.
Authors
Juan Pablo Vigneaux, Jennifer Cole
Abstract
Text-to-speech systems built on residual vector quantization often generate audio in two stages: a transformer backbone emits the first level of each frame's token stack, and a smaller depth model generates the rest autoregressively. We ask which components of such a model are responsible for prosodic intonation, more precisely for generating the English nuclear tune (for instance, the falling pitch melody that distinguishes an assertion from a question's rising melody), and whether interventions on those components can be used to control intonation. Prompt-based control mostly fails; only a final question mark is a reliable lever. Interchange interventions in CSM and S2 locate the tune: the backbone's hidden state carries it to the waveform through the depth model, and the first level of the stack carries little or none of it. Steering the backbone state with a difference-of-means vector computed on new material generated with a final period or question mark moves the tune, but the push toward an assertion degrades the voice. Steering vectors computed instead from the backbone states associated with human imitations of theoretically-specified tunes, whose text carries no punctuation, and applied inside the generation loop, move most recordings to the other condition in both systems (even when fitted without the speaker being steered), at a small cost in predicted naturalness. These imitation-derived vectors are aligned with the one obtained from the two punctuated conditions far beyond a shuffled-label null, so the backbone state of a punctuated prompt already represents intonation. The steered recordings broadly become falling or rising, according to the intent, but the edge tone moves more coherently than the pitch accent, and the target tune cannot be selected precisely beyond this broad tendency.