whisper-large-v3-turbo-anv-zul (CTranslate2, int8)
A CTranslate2 conversion of
dsfsi-anv/whisper-large-v3-turbo-anv-zul,
a Whisper large-v3-turbo fine-tune for Zulu trained by DSFSI on the
African Next Voices corpus. The weights are unchanged apart from the format
conversion and int8 quantization, for use with
faster-whisper.
All credit for the model goes to the original authors. Please cite and refer to
the source model
and the African Next Voices dataset.
Usage
Whisper has no isiZulu language token. The fine-tune decodes under the English
token (en), which Whisper's language detection narrowly picks for it, so pass
it explicitly to avoid detection drifting to another language. Turning off
condition_on_previous_text avoids repetition loops on long audio.
from faster_whisper import WhisperModel
model = WhisperModel("gyorilab/whisper-large-v3-turbo-anv-zul-ct2",
device="cpu", compute_type="int8")
segments, _ = model.transcribe("audio.wav", language="en",
condition_on_previous_text=False)
print(" ".join(segment.text for segment in segments))
Conversion
Converted from source revision 45ca90eca9c5e7b64510a59408b75ba4e5e7b713 with
CTranslate2 4.8.0:
ct2-transformers-converter --model dsfsi-anv/whisper-large-v3-turbo-anv-zul \
--revision 45ca90eca9c5e7b64510a59408b75ba4e5e7b713 \
--output_dir whisper-large-v3-turbo-anv-zul-ct2 \
--copy_files preprocessor_config.json --quantization int8
tokenizer.json is the source model's fast tokenizer, saved with
transformers, since the source repository ships only the slow tokenizer files.
License
MIT, as for the source model.
