whisper-large-v3-turbo-german-ct2

CTranslate2 (float16) conversion of primeline/whisper-large-v3-turbo-german for use with faster-whisper.

All credit for the model goes to primeLine — see the original model card for training details and German WER benchmarks.

Conversion

Converted from source revision 9e7012dad9247f07553aac8a99c4d50ad7c3d36d with:

ct2-transformers-converter \
    --model primeline/whisper-large-v3-turbo-german \
    --output_dir whisper-large-v3-turbo-german-ct2 \
    --quantization float16

tokenizer.json was generated with transformers.WhisperTokenizerFast (the source repo only ships the slow tokenizer files), and preprocessor_config.json was copied from the source repo.

Usage

from faster_whisper import WhisperModel

model = WhisperModel("ostwilkens/whisper-large-v3-turbo-german-ct2", device="cuda", compute_type="float16")
segments, info = model.transcribe("audio.wav", language="de")