CodeWithAhsan/whisper-medium-urdu-ggml

Whisper Medium Urdu — whisper.cpp GGML conversion

A whisper.cpp-ready GGML conversion of Abdul145/whisper-medium-urdu-custom (Apache-2.0), a Whisper-medium fine-tune for Urdu automatic speech recognition. All training credit belongs to the original author — this repo only re-packages the weights so they load directly in whisper.cpp-based tools without a Python conversion step.

Files

• ggml-medium-urdu.bin (~1.4 GB) — sha1 59769d590f62eeeb3bc3f5b82ce8c03b6e96831e

Converted with whisper.cpp's models/convert-h5-to-ggml.py.

Use with whisper.cpp

whisper-cli -m ggml-medium-urdu.bin -l ur -f audio.wav

Use with ossclip (word-timed Urdu captions, RTL layout included)

ossclip setup --model medium-urdu
ossclip produce video.mp4 --whisper-model medium-urdu --captions

The language (ur) is implied by the model; ossclip lays out right-to-left caption lines automatically and bundles Noto Nastaliq Urdu for rendering.

License

Apache-2.0, inherited from the base fine-tune. See the original model card for training details.