Troubleshooting

Troubleshooting

lwt not found (Windows)

python -m local_whisper_transcribe.cli transcribe meeting.mp4

Add Python Scripts to PATH — see Installation.

ffmpeg not detected

Restart terminal after winget install ffmpeg.

$env:Path += ";$env:LOCALAPPDATA\Microsoft\WinGet\Links"
ffmpeg -version

lwt also auto-detects winget ffmpeg paths.

Model not downloaded

lwt setup
lwt models download small

Diarization 403

Symptoms: 403, gated repo, not authorized

Fix:

1. Accept licenses for all 3 models — HuggingFace Diarization
2. Create Read token at https://huggingface.co/settings/tokens
3. lwt config set diarization.hf_token hf_YOUR_TOKEN
4. Retry

Skip diarization:

lwt transcribe meeting.mp4 --summarize

CUDA / cublas64_12.dll

lwt install cuda

Or force CPU:

lwt config set whisper.device cpu

See GPU and CUDA.

Ollama not running

lwt ollama status
lwt ollama pull llama3.2

Slow transcription

• Use smaller model: lwt config set whisper.model small
• Install GPU libs: lwt install cuda

Re-running after diarization failure

Whisper runs from scratch each time. To test diarization quickly:

lwt transcribe meeting.mp4 --diarize --model small

Use large-v3 only after HF setup is verified.