Indic Transcribe Core INT8
100% On-Device, Quantized INT8 ONNX Speech Recognition Package for 25 Indian Languages (Native Vernacular Focus)
This repository provides a modular, weight-quantized, Microsoft ONNX Runtime-compatible derivative of Bodhan AI's Indic-Transcribe-Core model, specifically optimized for private, high-fidelity native-script speech recognition in on-device mobile applications.
Model Lineage & Provenance
- Derivative Author: Community Contributor (2026)
- Direct Upstream Model: bodhan-ai/indic-transcribe-core
- Upstream Commit SHA:
4d29eeb7a0990de4a8febf9a4d5a9c5c61134a0d - Underlying Base Architecture: nvidia/canary-1b-v2
- Base Architecture License: Creative Commons Attribution 4.0 International (CC-BY-4.0)
- Upstream License: Indic Open Model License v1.0
This model is a quantized derivative of
bodhan-ai/indic-transcribe-core. It is specialized for maximum accuracy in native Indic scripts (such as เฌเฌกเฌผเฌฟเฌ, เคฆเฅเคตเคจเคพเคเคฐเฅ, เฎคเฎฎเฎฟเฎดเฏ, etc.) for formal documentation, news, legal, and educational voice transcription.
Quantization Specifications
- Quantization Type: Dynamic Activation + Static Weight INT8 (W8A8 Dynamic Linear)
- Quantized Projections: 530
nn.Linearlayers (289 in Encoder, 241 in Decoder) mapped to ONNXMatMulIntegeroperations - Weight Initializers: 1,060 INT8 weight tensors (
DataType 3) - Dynamic Activation Quantization: Performed via ONNX
DynamicQuantizeLinearnodes - Preserved High-Precision Paths: Convolution subsampling blocks, LayerNorm layers, Rotary Positional Embeddings, and residual additions remain strictly in FP32 (
DataType 1) - Footprint Comparison:
- Upstream FP32 Checkpoint: 4.55 GB (
model.safetensors) - Quantized INT8 ONNX (
encoder.onnx+decoder.onnx): 1.45 GB (1,570,095,829 bytes) - Storage Reduction: 68.1%
Supported Languages (25 Languages)
The Core model is optimized for high-accuracy native transcription across 25 languages:
- Official Languages:
as(Assamese),bn(Bengali),brx(Bodo),doi(Dogri),gu(Gujarati),hi(Hindi),kn(Kannada),ks(Kashmiri),kok(Konkani),mai(Maithili),ml(Malayalam),mni(Manipuri),mr(Marathi),ne(Nepali),or(Odia),pa(Punjabi),sa(Sanskrit),sat(Santali),sd(Sindhi),ta(Tamil),te(Telugu),ur(Urdu),en(English) - Low-Resource Varieties:
bho(Bhojpuri),bhb(Bhili)
Package Structure
indic-transcribe-core-int8-onnx/
โโโ README.md # Model Card (this document)
โโโ LICENSE # Full text of Indic Open Model License v1.0
โโโ LICENSE_DEED.md # Plain-language deed summary
โโโ NOTICE.md # Legal attribution for Bodhan AI and NVIDIA
โโโ config/
โ โโโ config.json # Canary FastConformer 32L/24L dimensions
โ โโโ tokenizer_config.json # Tokenizer layout & language prompt IDs
โ โโโ generation_config.json # Autoregressive generation parameters
โโโ model/
โ โโโ encoder.onnx # 32-layer FastConformer Encoder (1.04 GB)
โ โโโ decoder.onnx # 24-layer Transformer Decoder (427.5 MB)
โ โโโ bodhan_vocab.json # 7,152 token piece array for mobile decoding (84.8 KB)
โ โโโ tokenizer_multilingual.model # SentencePiece multilingual vocabulary
โ โโโ tokenizer_spl_tokens.model # SentencePiece special tokens vocabulary
โโโ metadata/
โโโ manifest.json # Versioned download manifest
โโโ checksums.sha256 # SHA-256 verification sums
Android App Consumption Contract
On-device mobile applications consumes this repository deterministically:
- Queries
metadata/manifest.json. - Verifies client app version compatibility (
minimum_app_version_code: 295). - Confirms available disk storage (> 2.0 GB free).
- Streams
model/encoder.onnx,model/decoder.onnx, andmodel/bodhan_vocab.jsondirectly into internal storage. - Computes streaming SHA-256 digests and validates against
manifest.json. - Atomically moves verified models into the internal app storage path:
/files/models/bodhan_core_int8/. - Instantiates ONNX Runtime CPU sessions in
CanaryAedEngine.kt.
Limitations & Usage Guidelines
- Native Script Target: Optimized for native Indic scripts. For mixed English-Hindi (Hinglish) transliteration, use Indic Transcribe Flex INT8 instead.
- Acoustic Conditions: Optimized for single-speaker conversational speech.
- Hardware Requirements: On-device INT8 CPU inference requires an ARM64-v8a processor with at least 4 GB total system RAM (2 GB free during inference).
License & Legal Terms
- Model Distribution: Released under the Indic Open Model License v1.0.
- Base Architecture: NVIDIA Canary-1B-v2 released under CC-BY-4.0.
- Commercial Use: Commercial on-device self-hosted inference is permitted under the Indic Open Model License v1.0. Products exceeding 500M+ MAU or $250M+ annual revenue require a separate commercial license from Bodhan AI.
- Hosting Restriction: This repository distributes files for local on-device client inference. Providing this model as a hosted third-party API service requires written approval from Bodhan AI.
- Attribution:
"Built with Indic-Transcribe-Core from Bodhan AI / AI4Bharat."
"Base model: Canary-1B-v2 developed by NVIDIA Corporation (CC-BY-4.0)."
Prohibited Uses (Section 10)
As mandated by Section 10 of the Indic Open Model License v1.0, this model and its derivatives must not be used for:
- Child sexual abuse material or content sexualizing minors.
- Conventional, chemical, biological, radiological, or nuclear weapons development.
- Mass surveillance or social-scoring systems.
- Disinformation campaigns or election manipulation.
- Automated decisions materially affecting legal rights without human oversight.
- Non-consensual voice cloning or deepfakes of real individuals.
- Robocalls, auto-dialers, or voice-phishing (vishing) schemes.
- AI companion products designed to simulate romantic intimacy or emotional dependency.
Citation & References
ibtex @misc{indictranscribe2026, title = {Indic-Transcribe: Built for the way India actually speaks}, author = {Bodhan AI, AI4Bharat}, year = {2026}, url = {https://bodhan.ai/research/blogs/indic-transcribe} }
- Upstream Model Card: bodhan-ai/indic-transcribe-core
- Base Architecture: nvidia/canary-1b-v2
- Upstream License: Indic Open Model License v1.0