facebookresearch/fairseq

Is same base model used for MMS-ASR and MMS-TTS (like AudioPaLM)?

Open

#5,221 opened on Jun 27, 2023

View on GitHub
 (1 comment) (0 reactions) (0 assignees)Python (29,107 stars) (6,224 forks)batch import
enhancementhelp wantedneeds triage

Description

Google introduce AudioPaLM, a large language model for speech understanding and generation. AudioPaLM fuses text-based and speech-based language models, PaLM-2 [Anil et al., 2023] and AudioLM [Borsos et al., 2022], into a unified multimodal architecture that can process and generate text and speech with applications including speech recognition and speech-to-speech translation.

How about the MMS? I found fine-tuning MMS ASR based on pretrained base model mms-1b, but I can not find for TTS. Is the same base model mms-1b used for MMS-TTS? How can I fine-tuning or add new language for TTS?

Contributor guide

Tech stack
pythonpytorch
Domain
machine learningai
Issue type
research
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-3 hours
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
clear
Prerequisites
basic knowledge of speech modelsfamiliarity with fairseq
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
30
Research direction
The issue asks whether the same base model (mms 1b) is used for both ASR and TTS in MMS. To answer, review the MMS paper (https://arxiv.org/abs/2305.13516) and the model release in the repository. Check the fairseq documentation for TTS fine tuning scripts. Look for any separate TTS checkpoints or configs. If none exist, note that TTS support may not be publicly available. Compare with AudioPaLM architecture as mentioned in the issue.