KB-Whisper Large for faster-whisper
This repository contains only the CTranslate2 files from
KBLab/kb-whisper-large, a
Swedish Whisper large-v3 model created by KBLab at the National Library of
Sweden. The files were copied unchanged from upstream revision
d5d5984b4d8f7c4847a8ea203f1976285fb28300.
The upstream model card contains the training data, evaluation results, limitations, acknowledgements, and citation. KBLab reports that the model was trained on more than 50,000 hours of Swedish speech.
Usage
from faster_whisper import WhisperModel
model = WhisperModel("aTrain-core/KB-WhisperSwedish", device="cpu", compute_type="int8")
segments, info = model.transcribe("audio.mp3", language="sv", word_timestamps=True)
for segment in segments:
print(f"[{segment.start:.2f}s -> {segment.end:.2f}s] {segment.text}")
The files were tested with faster-whisper==1.2.1 on CPU using int8.
License and attribution
The upstream model is distributed under the Apache License 2.0. See
LICENSE. KB-Whisper is a product of KBLab at the National Library
of Sweden. Please use the citation provided in the
upstream model card.
- Downloads last month
- 30