Zubaan model mirror
Byte-identical copies of the third-party speech models the Zubaan app (by Arinverse, https://arinverse.github.io/) downloads, hosted here so the app does not depend on any upstream account staying available. No file was modified from its source except where noted.
| Folder | Original source | License |
|---|---|---|
| en/ | k2-fsa icefall Zipformer, via csukuangfj/sherpa-onnx-streaming-zipformer-en-2023-06-26 | Apache-2.0 |
| bn/ | alphacep/vosk-model-small-streaming-bn, via csukuangfj2/sherpa-onnx-streaming-zipformer-bn-vosk-2026-02-09 | Apache-2.0 |
| hi/ | AI4Bharat IndicConformer, ONNX export OpenVoiceOS/ai4bharat-indicconformer-hi-onnx | MIT |
| es/ | nvidia/stt_es_fastconformer_hybrid_large_pc, ONNX + int8 by csukuangfj (sherpa-onnx) | CC-BY-4.0 |
| pt/ | neongeckocom/stt_pt_citrinet_512_gamma_0_25, ONNX export by OpenVoiceOS | BSD-3-Clause |
| whisper/ | OpenAI Whisper, ggml conversion ggerganov/whisper.cpp | MIT |
| shared/ | Silero VAD, via k2-fsa/sherpa-onnx releases | MIT |
| punctuation/ | 1-800-BAD-CODE/punct_cap_seg_47_language, int8-quantised (modified) | Apache-2.0 |
Spanish model attribution (CC BY 4.0): "stt_es_fastconformer_hybrid_large_pc" by NVIDIA Corporation, https://creativecommons.org/licenses/by/4.0/ — changes: exported to ONNX and quantised to int8 by the sherpa-onnx project.
Portuguese model attribution (CC BY 4.0): fine-tuned by NeonGecko.com Inc. from "stt_en_citrinet_512_gamma_0_25" by NVIDIA Corporation, https://creativecommons.org/licenses/by/4.0/ — changes: fine-tuned on Portuguese speech by NeonGecko; exported to ONNX by OpenVoiceOS.
All credit for training and conversion belongs to the original authors above.