Speech Synthesis From Continuous Features Using Per-Token Latent Diffusion

Arnon Turetzky; Avihu Dekel; Nimrod Shabtay; Slava Shechtman; David Haws; Hagai Aronowitz; Ron Hoory; Yossi Adi

doi:10.1109/asru65441.2025.11434698

Speech Synthesis From Continuous Features Using Per-Token Latent Diffusion

Arnon Turetzky^*
, Avihu Dekel
, Nimrod Shabtay
, Slava Shechtman
, David Haws
, Hagai Aronowitz
, Ron Hoory
, Yossi Adi

^*Corresponding author for this work

The Rachel and Selim Benin School of Engineering and Computer Science

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

We present SALAD, a zero-shot text-to-speech (TTS) autoregressive model operating over continuous speech representations. SALAD utilizes a per-token diffusion process to refine and predict continuous representations for the next time step. We compare our approach against a discrete variant of SALAD as well as publicly available zero-shot TTS systems, and conduct a comprehensive analysis of discrete versus continuous modeling techniques. Our results show that SALAD achieves superior intelligibility while matching the speech quality and speaker similarity of ground-truth audio.

Original language	English
Title of host publication	ASRU 2025 - 2025 IEEE Automatic Speech Recognition and Understanding Workshop
Publisher	Institute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)	9798331544263
DOIs	https://doi.org/10.1109/asru65441.2025.11434698
State	Published - 2025
Event	2025 IEEE Automatic Speech Recognition and Understanding Workshop, ASRU 2025 - Honolulu, United States Duration: 6 Dec 2025 → 10 Dec 2025

Publication series

Name	ASRU 2025 - 2025 IEEE Automatic Speech Recognition and Understanding Workshop

Conference

Conference	2025 IEEE Automatic Speech Recognition and Understanding Workshop, ASRU 2025
Country/Territory	United States
City	Honolulu
Period	6/12/25 → 10/12/25

Bibliographical note

Publisher Copyright:
© 2025 IEEE.

Access to Document

10.1109/asru65441.2025.11434698

Cite this

Turetzky, A., Dekel, A., Shabtay, N., Shechtman, S., Haws, D., Aronowitz, H., Hoory, R., & Adi, Y. (2025). Speech Synthesis From Continuous Features Using Per-Token Latent Diffusion. In ASRU 2025 - 2025 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU 2025 - 2025 IEEE Automatic Speech Recognition and Understanding Workshop). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/asru65441.2025.11434698

@inproceedings{88d204cb30a549c6bfbff9828d3d1dd8,

title = "Speech Synthesis From Continuous Features Using Per-Token Latent Diffusion",

abstract = "We present SALAD, a zero-shot text-to-speech (TTS) autoregressive model operating over continuous speech representations. SALAD utilizes a per-token diffusion process to refine and predict continuous representations for the next time step. We compare our approach against a discrete variant of SALAD as well as publicly available zero-shot TTS systems, and conduct a comprehensive analysis of discrete versus continuous modeling techniques. Our results show that SALAD achieves superior intelligibility while matching the speech quality and speaker similarity of ground-truth audio.",

author = "Arnon Turetzky and Avihu Dekel and Nimrod Shabtay and Slava Shechtman and David Haws and Hagai Aronowitz and Ron Hoory and Yossi Adi",

note = "Publisher Copyright: {\textcopyright} 2025 IEEE.; 2025 IEEE Automatic Speech Recognition and Understanding Workshop, ASRU 2025 ; Conference date: 06-12-2025 Through 10-12-2025",

year = "2025",

doi = "10.1109/asru65441.2025.11434698",

language = "אנגלית",

series = "ASRU 2025 - 2025 IEEE Automatic Speech Recognition and Understanding Workshop",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

booktitle = "ASRU 2025 - 2025 IEEE Automatic Speech Recognition and Understanding Workshop",

address = "ארצות הברית",

}

Turetzky, A, Dekel, A, Shabtay, N, Shechtman, S, Haws, D, Aronowitz, H, Hoory, R & Adi, Y 2025, Speech Synthesis From Continuous Features Using Per-Token Latent Diffusion. in ASRU 2025 - 2025 IEEE Automatic Speech Recognition and Understanding Workshop. ASRU 2025 - 2025 IEEE Automatic Speech Recognition and Understanding Workshop, Institute of Electrical and Electronics Engineers Inc., 2025 IEEE Automatic Speech Recognition and Understanding Workshop, ASRU 2025, Honolulu, United States, 6/12/25. https://doi.org/10.1109/asru65441.2025.11434698

Speech Synthesis From Continuous Features Using Per-Token Latent Diffusion. / Turetzky, Arnon; Dekel, Avihu; Shabtay, Nimrod et al.
ASRU 2025 - 2025 IEEE Automatic Speech Recognition and Understanding Workshop. Institute of Electrical and Electronics Engineers Inc., 2025. (ASRU 2025 - 2025 IEEE Automatic Speech Recognition and Understanding Workshop).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Speech Synthesis From Continuous Features Using Per-Token Latent Diffusion

AU - Turetzky, Arnon

AU - Dekel, Avihu

AU - Shabtay, Nimrod

AU - Shechtman, Slava

AU - Haws, David

AU - Aronowitz, Hagai

AU - Hoory, Ron

AU - Adi, Yossi

PY - 2025

Y1 - 2025

N2 - We present SALAD, a zero-shot text-to-speech (TTS) autoregressive model operating over continuous speech representations. SALAD utilizes a per-token diffusion process to refine and predict continuous representations for the next time step. We compare our approach against a discrete variant of SALAD as well as publicly available zero-shot TTS systems, and conduct a comprehensive analysis of discrete versus continuous modeling techniques. Our results show that SALAD achieves superior intelligibility while matching the speech quality and speaker similarity of ground-truth audio.

AB - We present SALAD, a zero-shot text-to-speech (TTS) autoregressive model operating over continuous speech representations. SALAD utilizes a per-token diffusion process to refine and predict continuous representations for the next time step. We compare our approach against a discrete variant of SALAD as well as publicly available zero-shot TTS systems, and conduct a comprehensive analysis of discrete versus continuous modeling techniques. Our results show that SALAD achieves superior intelligibility while matching the speech quality and speaker similarity of ground-truth audio.

UR - https://www.scopus.com/pages/publications/105036554833

U2 - 10.1109/asru65441.2025.11434698

DO - 10.1109/asru65441.2025.11434698

M3 - ???researchoutput.researchoutputtypes.contributiontobookanthology.conference???

AN - SCOPUS:105036554833

T3 - ASRU 2025 - 2025 IEEE Automatic Speech Recognition and Understanding Workshop

BT - ASRU 2025 - 2025 IEEE Automatic Speech Recognition and Understanding Workshop

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 2025 IEEE Automatic Speech Recognition and Understanding Workshop, ASRU 2025

Y2 - 6 December 2025 through 10 December 2025

ER -

Turetzky A, Dekel A, Shabtay N, Shechtman S, Haws D, Aronowitz H et al. Speech Synthesis From Continuous Features Using Per-Token Latent Diffusion. In ASRU 2025 - 2025 IEEE Automatic Speech Recognition and Understanding Workshop. Institute of Electrical and Electronics Engineers Inc. 2025. (ASRU 2025 - 2025 IEEE Automatic Speech Recognition and Understanding Workshop). doi: 10.1109/asru65441.2025.11434698

Speech Synthesis From Continuous Features Using Per-Token Latent Diffusion

Abstract

Publication series

Conference

Bibliographical note

Access to Document

Other files and links

Fingerprint

Cite this