NewSraVaani 1.0 · 65 languages

From India’s Voices
To Open Source Speech Models.

Project Vaani began as a dataset of 31,255 hours of natural speech from 165 districts. That corpus now powers 20 open source speech models: multilingual and monolingual speech recognition, speech synthesis, and spoken language identification. Every one of them is released on Hugging Face under the MIT licence, free to download, fine-tune and deploy.

20Open source models
60+Languages & dialects
31,255 hrPretraining speech
11 MAudio–image pairs

Flagship model

SraVaani 1.0

A 430M-parameter FastConformer that recognises speech across 65 Indian languages and dialects, including Tulu, Wancho, Kokborok and Bearybashe, which no commercial ASR system covers at all. Pretrained from scratch on Vaani, then grounded against 11 million audio–image pairs.

Interactive

Try the models in your browser.

Inference runs on Hugging Face Spaces, not on this server. The demos below are the official ARTPARK-IISc Spaces embedded in the page. Speak or upload a clip to transcribe it, or type text to hear it synthesised.

Vaani demos
Open on Hugging Face ↗
Demos hosted by ARTPARK-IISc on Hugging Face Spaces. If a Space is asleep, give it a few seconds to spin up or open it directly on Hugging Face. The streaming SraVaani live Space requires sign-in and so is linked rather than embedded.

Model catalogue

Organised by what you need it to do.

Start with the task: recognise speech, synthesise it, or identify the language being spoken. Within speech recognition, choose a multilingual model if you do not know the input language in advance, or a monolingual checkpoint if you do and want something smaller and sharper.

Automatic Speech Recognition

Multilingual2 models

One model, many languages. Handles code-mixed and unknown-language input without needing to be told what is being spoken. That makes it the right default for most applications.

SraVaani 1.0

Our flagship. A 430M-parameter FastConformer with a hybrid TDT–CTC decoder, pretrained from scratch on Vaani and fine-tuned across 65 Indian languages and dialects.

65 languages430M paramsFP16 · ~900 MBMIT

SraVaani 0.5-live

The streaming sibling: a lighter, low-latency model that emits the transcript while the person is still speaking. Built for voice interfaces, live captioning and call-centre use.

Real-timeStreamingLow latency

Vaani-FastConformer

Monolingual5 models

Single-language FastConformer checkpoints trained on Vaani. Smaller and faster than the multilingual model, and the right pick when you already know the input language and want to minimise compute.

ModelLanguageArchitectureTask
Vaani-FastConformer-HindiHindiFastConformerASROpen ↗
Vaani-FastConformer-KannadaKannadaFastConformerASROpen ↗
Vaani-FastConformer-TeluguTeluguFastConformerASROpen ↗
Vaani-FastConformer-MalayalamMalayalamFastConformerASROpen ↗
Vaani-FastConformer-OdiaOdiaFastConformerASROpen ↗

View the full collection on Hugging Face ↗

Whisper fine-tuned on Vaani

Monolingual11 models

OpenAI Whisper checkpoints fine-tuned on Vaani data, released across the full size ladder from tiny (37.8M) through large-v3 (2B). Pick a size to match your latency and hardware budget: tiny and small run on CPU or edge devices, large-v3 gives the best accuracy on a GPU.

ModelLanguageSizeParams
whisper-large-v3-vaani-hindiHindilarge-v32BOpen ↗
whisper-large-v3-vaani-teluguTelugularge-v32BOpen ↗
whisper-large-v3-vaani-odiaOdialarge-v32BOpen ↗
whisper-medium-vaani-hindiHindimedium0.8BOpen ↗
whisper-medium-vaani-kannadaKannadamedium0.8BOpen ↗
whisper-medium-vaani-teluguTelugumedium0.8BOpen ↗
whisper-medium-vaani-bengaliBengalimedium0.8BOpen ↗
whisper-small-vaani-hindiHindismall0.2BOpen ↗
whisper-small-vaani-kannadaKannadasmall0.2BOpen ↗
whisper-small-vaani-tuluTulusmall0.2BOpen ↗
whisper-tiny-vaani-hindiHinditiny37.8MOpen ↗

Speech Synthesis

Multilingual1 model

Turning text back into natural Indian-language speech: the other half of a voice interface.

DhVaani 0.5

A compact ~0.1B parameter zero-shot TTS model covering 27 Indian languages, trained on curated Vaani audio. Use a built-in female or male voice, or clone a new one from a reference clip. Aimed at accessibility tooling, IVR systems and read-aloud content in languages mainstream TTS has never served.

27 languages0.1B paramsZero-shotVoice cloning

Spoken Language Identification

Multilingual1 model

The routing layer of a multilingual pipeline. Work out what is being spoken before deciding which model should transcribe it.

Vaani-LID v0

Identifies the spoken language directly from raw audio across 42 Indian languages, a scale of coverage that opens up genuinely new territory for Indic speech pipelines. A 0.8B-parameter feature-extraction model.

42 languages0.8B paramsFeature extraction

Complete index

All 20 models, in one table.

Filter by task or by language coverage.

TaskCoverage
ModelTaskCoverageFamilyParamsLanguagesDetails
SraVaani-1.0ASRMultilingualFastConformer TDT–CTC430M65Model page →
SraVaani-0.5-liveASRMultilingualStreaming–MultiOpen ↗
DhVaani-0.5TTSMultilingualZero-shot TTS0.1B27Open ↗
Vaani-LID_v0LIDMultilingualFeature extraction0.8B42Open ↗
Vaani-FastConformer-HindiASRMonolingualFastConformer–HindiOpen ↗
Vaani-FastConformer-KannadaASRMonolingualFastConformer–KannadaOpen ↗
Vaani-FastConformer-TeluguASRMonolingualFastConformer–TeluguOpen ↗
Vaani-FastConformer-MalayalamASRMonolingualFastConformer–MalayalamOpen ↗
Vaani-FastConformer-OdiaASRMonolingualFastConformer–OdiaOpen ↗
whisper-large-v3-vaani-hindiASRMonolingualWhisper large-v32BHindiOpen ↗
whisper-large-v3-vaani-teluguASRMonolingualWhisper large-v32BTeluguOpen ↗
whisper-large-v3-vaani-odiaASRMonolingualWhisper large-v32BOdiaOpen ↗
whisper-medium-vaani-hindiASRMonolingualWhisper medium0.8BHindiOpen ↗
whisper-medium-vaani-kannadaASRMonolingualWhisper medium0.8BKannadaOpen ↗
whisper-medium-vaani-teluguASRMonolingualWhisper medium0.8BTeluguOpen ↗
whisper-medium-vaani-bengaliASRMonolingualWhisper medium0.8BBengaliOpen ↗
whisper-small-vaani-hindiASRMonolingualWhisper small0.2BHindiOpen ↗
whisper-small-vaani-kannadaASRMonolingualWhisper small0.2BKannadaOpen ↗
whisper-small-vaani-tuluASRMonolingualWhisper small0.2BTuluOpen ↗
whisper-tiny-vaani-hindiASRMonolingualWhisper tiny37.8MHindiOpen ↗
Showing all 20 models. Weights, model cards and evaluation results are hosted on Hugging Face under ARTPARK-IISc.

Training data

The data behind the models.

Everything the models were trained and measured on is public, released before the models themselves were.

Vaani corpus

The full multimodal dataset: 22.6M rows of speech and images from 165 districts, released under CC-BY-4.0.

Transcription subset

The transcribed portion used for supervised fine-tuning across the language families.

Vaani Benchmark v1.0

The evaluation set behind the published WER figures, with a public leaderboard on Hugging Face.

Open source, end to end.

Every model on this page is open source, trained on data we open-sourced first. Download the weights, fine-tune them, ship them commercially, and tell us what you build.