Whisper-Hindi2Hinglish-Prime (ggml for whisper.cpp)
This is a whisper.cpp / ggml (f16) conversion of Oriserve/Whisper-Hindi2Hinglish-Prime, a fine-tune of OpenAI Whisper large-v3 that transcribes Hindi / Hinglish speech into Roman-script Hinglish.
- Original model: Oriserve, Apache-2.0. All credit for training goes to Oriserve.
- Change: converted to ggml with whisper.cpp
models/convert-h5-to-ggml.py. The bf16 weights were cast to float, and the result is stored as f16. No retraining or other weight changes. - Used by: Caption Bunny (Adobe Premiere Pro caption panel) as its offline Hinglish engine.
Usage
whisper-cli -m ggml-hinglish-prime-large-v3.bin -f audio_16k.wav -l hi -nfa -dtw large.v3
License
Apache License 2.0, the same as the original model. See the original model card for its details, limitations and evaluation.