Stochastic Pitch Prediction Improves the Diversity and Naturalness of Speech in Glow-TTS - Inria - Institut national de recherche en sciences et technologies du numérique Access content directly
Conference Papers Year : 2023

Stochastic Pitch Prediction Improves the Diversity and Naturalness of Speech in Glow-TTS

Abstract

Flow-based generative models are widely used in text-to-speech (TTS) systems to learn the distribution of audio features (e.g., Mel-spectrograms) given the input tokens and to sample from this distribution to generate diverse utterances. However, in the zero-shot multi-speaker TTS scenario, the generated utterances lack diversity and naturalness. In this paper, we propose to improve the diversity of utterances by explicitly learning the distribution of fundamental frequency sequences (pitch contours) of each speaker during training using a stochastic flow-based pitch predictor, then conditioning the model on generated pitch contours during inference. The experimental results demonstrate that the proposed method yields a significant improvement in the naturalness and diversity of speech generated by a Glow-TTS model that uses explicit stochastic pitch prediction, over a Glow-TTS baseline and an improved Glow-TTS model that uses a stochastic duration predictor.
Fichier principal
Vignette du fichier
Stochastic_Pitch_Prediction_for_Improving_the_Diversity_and_Naturalness_in_GlowTTS.pdf (281.52 Ko) Télécharger le fichier
Origin Files produced by the author(s)

Dates and versions

hal-04108825 , version 1 (28-05-2023)

Licence

Identifiers

  • HAL Id : hal-04108825 , version 1

Cite

Sewade Ogun, Vincent Colotte, Emmanuel Vincent. Stochastic Pitch Prediction Improves the Diversity and Naturalness of Speech in Glow-TTS. InterSpeech 2023, Aug 2023, Dublin, Ireland. ⟨hal-04108825⟩
57 View
45 Download

Share

Gmail Mastodon Facebook X LinkedIn More