Text-to-Speech for Low-Resource Languages
Building speech synthesis for languages with little recorded speech, from massively multilingual systems to regional voices.
Tier Grades
Tier A: Frontier AI labs
feasible
Meta built synthesis for 1,107 languages in MMS with a 16-author team.
Tier B: Indonesian AI labs
feasible
AI4Bharat shows a regional lab can train a 20-language TTS on 1,806 hours, a model for Indonesian and regional languages.
Tier C: Institution units such as ITB
feasible
Recording and curating regional speech is institution-scale work, as the Indic Parler-TTS data shows.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
stretch
Delivery teams can use existing Indonesian voices but rarely have the data to build new regional languages.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeNot in the sources read: MMS and Indic Parler-TTS give no training compute.
- DataMMS: speech synthesis models for 1,107 languages from readings of religious texts; Indic Parler-TTS: 1,806 hours of Indic and English speech for 20 Indic languages and English.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 16 authors on MMS.
- Time horizonMonths.
- Shelf-life signalsCommercial TTS covering Indonesian regional languages such as Javanese and Sundanese at high quality
- Pipeline roleTier A releases multilingual synthesis; Tier C and Tier B record and train Indonesian regional voices; Tier D uses them in public service and education pilots.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Linked models
Links
R&D TopicText-to-Speech for Low-Resource Languages
Models
ai4bharat indic-parler-ttsAI4Bharat (IIT Madras), speech synthesis
MMSMeta, speech recognition
Eleven Multilingual v2ElevenLabs, speech synthesisEleven v4ElevenLabs, speech synthesis+2 more
ai4bharat indic-parler-ttsAI4Bharat (IIT Madras), speech synthesisMMSMeta, speech recognitionEleven Multilingual v2ElevenLabs, speech synthesisEleven v4ElevenLabs, speech synthesis+2 moreUse Casesnone recorded
Who Runs Itnone recorded
Models 6
MMSMeta, speech recognitionexemplarEleven v4ElevenLabs, speech synthesisproduced by
sanoTTSAmpixa, speech synthesisproduced byai4bharat indic-parler-ttsAI4Bharat (IIT Madras), speech synthesisexemplarEleven Multilingual v2ElevenLabs, speech synthesisproduced by
Magpie TTS Multilingual 357MNVIDIA, speech synthesisproduced byPapers
- Learning to Speak Fluently in a Foreign Language: Multilingual Speech Synthesis and Cross-Language Voice CloningYu Zhang et al., Interspeech 2019 (per arXiv comment), 2019 founding
- XTTS: a Massively Multilingual Zero-Shot Text-to-Speech ModelEdresson Casanova et al., INTERSPEECH 2024 (per arXiv comment), 2024 state of the art
- Scaling Speech Technology to 1,000+ LanguagesVineel Pratap et al., arXiv preprint, 2023 state of the art
- YourTTS: Towards Zero-Shot Multi-Speaker TTS and Zero-Shot Voice Conversion for everyoneEdresson Casanova et al., Proceedings of the 39th International Conference on Machine Learning, PMLR 162:2, 2021 state of the art
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2305.13516 https://arxiv.org/abs/2305.13516
- huggingface.co/ai4bharat/indic-parler-tts https://huggingface.co/ai4bharat/indic-parler-tts
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Text-to-Speech for Low-Resource Languages", https://aurion.semai.id/topics/tts-low-resource-languages/, records of October 2026.