q-henric/nb-whisper-large-ct2

NB-Whisper Large (CTranslate2 Format)

This is a CTranslate2 conversion of NbAiLab/nb-whisper-large for use with WhisperX and faster-whisper.

Model Details

• Original Model: NbAiLab/nb-whisper-large
• Developed by: NB AI-Lab (National Library of Norway)
• Model Size: Large (1550M parameters)
• Languages: Norwegian (Bokmål, Nynorsk), English
• License: Apache 2.0 (same as original)

Why This Repository?

The original NbAiLab/nb-whisper-large model has CTranslate2 files in a /ct2 subfolder. WhisperX and faster-whisper expect CT2 files at the repository root. This repository provides the CT2 files at root level for direct compatibility.

Usage with WhisperX

import whisperx

model = whisperx.load_model(
    "q-henric/nb-whisper-large-ct2",
    device="cpu",  # or "cuda"
    compute_type="int8"
)

audio = whisperx.load_audio("audio.mp3")
result = model.transcribe(audio, language="no")

Usage with faster-whisper

from faster_whisper import WhisperModel

model = WhisperModel("q-henric/nb-whisper-large-ct2", device="cpu", compute_type="int8")
segments, info = model.transcribe("audio.mp3", language="no")

Files

• config.json - CTranslate2 model configuration
• model.bin - CTranslate2 model weights (~6.17 GB)
• vocabulary.json - Tokenizer vocabulary