FastPitchFormant: Source-filter based Decomposed Modeling for Speech Synthesis

TL;DR

FastPitchFormant introduces a source-filter based neural TTS model that improves prosody control and speech quality by separately modeling text and acoustic features, reducing pitch-shift artifacts and speaker deformation.

Contribution

It presents a novel feed-forward Transformer TTS model based on source-filter theory with parallel feature handling to enhance prosody and quality.

Findings

Reduces audio quality degradation in large pitch-shift synthesis.

Mitigates speaker characteristic deformation during prosody modification.

Separately models text and acoustic features for better control.

Abstract

Methods for modeling and controlling prosody with acoustic features have been proposed for neural text-to-speech (TTS) models. Prosodic speech can be generated by conditioning acoustic features. However, synthesized speech with a large pitch-shift scale suffers from audio quality degradation, and speaker characteristics deformation. To address this problem, we propose a feed-forward Transformer based TTS model that is designed based on the source-filter theory. This model, called FastPitchFormant, has a unique structure that handles text and acoustic features in parallel. With modeling each feature separately, the tendency that the model learns the relationship between two features can be mitigated.