pith. sign in

arxiv: 2605.29859 · v1 · pith:TDAC27ZWnew · submitted 2026-05-28 · 📡 eess.AS · cs.CL

MELD: Mel-Spectrogram-Based Speech Language Modeling with Discrete Latent Variables

classification 📡 eess.AS cs.CL
keywords languagespeechautoregressivediscretedownstreamencoderslatentmel-spectrogram-based
0
0 comments X
read the original abstract

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.

This paper has not been read by Pith yet.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.