Whisper large-v2 model for CTraslate2

This repository cotais the coversio of opeai/whisper-large-v2 to the CTraslate2 model format.

This model ca be used i CTraslate2 or projects based o CTraslate2 such as faster-whisper.

Example

from faster_whisper import WhisperModel

model = WhisperModel("large-v2")

segmets, ifo = model.trascribe("audio.mp3")
for segmet i segmets:
    prit("[%.2fs -> %.2fs] %s" % (segmet.start, segmet.ed, segmet.text))

Coversio details

The origial model was coverted with the followig commad:

ct2-trasformers-coverter --model opeai/whisper-large-v2 --output_dir faster-whisper-large-v2 \
    --copy_files tokeizer.jso --quatizatio float16

Note that the model weights are saved i FP16. This type ca be chaged whe the model is loaded usig the compute_type optio i CTraslate2.

More iformatio

For more iformatio about the origial model, see its model card.

Whisper large-v2 model for CTranslate2 This repository contains the conversion of openai/whisper-lar