Saved in:
Bibliographic Details
Main Authors: Yeh, Sung-Lin, Zhou, Wei, Keren, Gil, Le, Duc, Meng, Zhong, Tang, Hao, Mahadeokar, Jay, Kalinli, Ozlem, Mourachko, Alexandre
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.29859
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917543816462336
author Yeh, Sung-Lin
Zhou, Wei
Keren, Gil
Le, Duc
Meng, Zhong
Tang, Hao
Mahadeokar, Jay
Kalinli, Ozlem
Mourachko, Alexandre
author_facet Yeh, Sung-Lin
Zhou, Wei
Keren, Gil
Le, Duc
Meng, Zhong
Tang, Hao
Mahadeokar, Jay
Kalinli, Ozlem
Mourachko, Alexandre
contents Recent speech language models rely on encoders that are optimized separately from autoregressive models. Since these encoders are unaware of the downstream objectives, the extracted representations may not be optimal for downstream tasks. To address this limitation, we introduce a discrete latent variable model on mel spectrograms that jointly optimizes the encoder and the speech language model. Joint optimization not only brings improvements over codec-based and other mel-spectrogram-based baselines on zero-shot Text-to-Speech (TTS) and Speech-to-Text (STT) tasks, but also effectively alleviates common issues in autoregressive mel-spectrogram modeling, such as prolonged silence generation and word omissions.
format Preprint
id arxiv_https___arxiv_org_abs_2605_29859
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle MELD: Mel-Spectrogram-Based Speech Language Modeling with Discrete Latent Variables
Yeh, Sung-Lin
Zhou, Wei
Keren, Gil
Le, Duc
Meng, Zhong
Tang, Hao
Mahadeokar, Jay
Kalinli, Ozlem
Mourachko, Alexandre
Audio and Speech Processing
Computation and Language
Recent speech language models rely on encoders that are optimized separately from autoregressive models. Since these encoders are unaware of the downstream objectives, the extracted representations may not be optimal for downstream tasks. To address this limitation, we introduce a discrete latent variable model on mel spectrograms that jointly optimizes the encoder and the speech language model. Joint optimization not only brings improvements over codec-based and other mel-spectrogram-based baselines on zero-shot Text-to-Speech (TTS) and Speech-to-Text (STT) tasks, but also effectively alleviates common issues in autoregressive mel-spectrogram modeling, such as prolonged silence generation and word omissions.
title MELD: Mel-Spectrogram-Based Speech Language Modeling with Discrete Latent Variables
topic Audio and Speech Processing
Computation and Language
url https://arxiv.org/abs/2605.29859