We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.SD

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Sound

Title: Guided-TTS: A Diffusion Model for Text-to-Speech via Classifier Guidance

Abstract: We propose Guided-TTS, a high-quality text-to-speech (TTS) model that does not require any transcript of target speaker using classifier guidance. Guided-TTS combines an unconditional diffusion probabilistic model with a separately trained phoneme classifier for classifier guidance. Our unconditional diffusion model learns to generate speech without any context from untranscribed speech data. For TTS synthesis, we guide the generative process of the diffusion model with a phoneme classifier trained on a large-scale speech recognition dataset. We present a norm-based scaling method that reduces the pronunciation errors of classifier guidance in Guided-TTS. We show that Guided-TTS achieves a performance comparable to that of the state-of-the-art TTS model, Grad-TTS, without any transcript for LJSpeech. We further demonstrate that Guided-TTS performs well on diverse datasets including a long-form untranscribed dataset.
Comments: 15 pages, 5 figures, ICML'2022
Subjects: Sound (cs.SD); Artificial Intelligence (cs.AI); Audio and Speech Processing (eess.AS)
Cite as: arXiv:2111.11755 [cs.SD]
  (or arXiv:2111.11755v4 [cs.SD] for this version)

Submission history

From: Sungwon Kim [view email]
[v1] Tue, 23 Nov 2021 10:05:05 GMT (677kb,D)
[v2] Tue, 7 Dec 2021 09:13:13 GMT (678kb,D)
[v3] Sat, 29 Jan 2022 01:34:24 GMT (590kb,D)
[v4] Fri, 10 Jun 2022 15:01:34 GMT (593kb,D)

Link back to: arXiv, form interface, contact.