Research map / Large language models / Language, safety and interpretability
Scaling Speech Technology to 1,000+ Languages
Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi and 4 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| pre-train on 1,406 languages (MMS) instead of 128 languages (XLS-R), 300M size | wav2vec 2.0 ASR · FLEURS-61 | CER: 0.6 lower CER | Improved |
| pre-train on 1,406 languages (MMS) instead of 128 languages (XLS-R), 1B size | wav2vec 2.0 ASR · FLEURS-61 | CER: 0.7 CER difference | Improved |
| scale parameters from 300M to 1B | XLS-R · FLEURS-61 | CER: 3.2 CER decrease | Improved |
| scale parameters from 300M to 1B | MMS-lab · FLEURS-61 | CER: 3.0 CER improvement | Improved |
| scale dense ASR from 61 to 1,107 languages | dense MMS ASR · FLEURS-61 | CER: +5.1 CER | Hurt |
| scale dense ASR from 61 to 1,107 languages | dense MMS ASR · CV-49 | CER: +2.1 CER | Hurt |
| add language-specific adapters and heads (LSAH) (ablation) | MMS ASR scaled to 1,107 languages · FLEURS-61 | CER degradation: 5.1 -> 0.4 | Improved |
| add language-specific adapters and heads (LSAH) (ablation) | MMS ASR scaled to 1,107 languages · CV-49 | CER degradation: 2.1 -> 0.2 | Improved |
| use MMS-lab CTC model with n-gram LM instead of Whisper | MMS ASR · FLEURS-54 | WER: 58% relative WER reduction | Improved |
| scale LID from 126 to 4,017 languages | MMS LID · BABEL | accuracy: -3.6% | Hurt |
| scale LID from 126 to 4,017 languages | MMS LID · FLEURS | accuracy: -0.3% | No clear effect |
| train LID on MMS-lab-U+unlab instead of in-domain VoxLingua-107 | MMS LID · FLEURS (72 languages) | accuracy: -2.1% | Hurt |
Closest papers by meaning
- WorldSpeech: A Multilingual Speech Corpus from Around the World (NeurIPS 2026 · similarity 0.86)
- SwitchLingua: The First Large-Scale Multilingual and Multi-Ethnic Code-Switching Dataset (NeurIPS 2025 · similarity 0.85)
- OWLS: Scaling Laws for Multilingual Speech Recognition and Translation Models (ICML 2025 · similarity 0.84)
- WAXAL: A Large-Scale Multilingual African Language Speech Corpus (NeurIPS 2026 · similarity 0.84)
- Pangea: A Fully Open Multilingual Multimodal LLM for 39 Languages (ICLR 2025 · similarity 0.84)
- LAMA-UT: Language Agnostic Multilingual ASR Through Orthography Unification and Language-Specific Transliteration (AAAI 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
