api-whisper-webui-git

name	api-whisper-webui-git
description	Transcribe local audio through Whisper-WebUI's HTTPS REST or Gradio API and retain diarized results.

Whisper-WebUI API

Clients

Use clients/transcribe.py for file transcription. It auto-detects the documented polling REST API from openapi.json; otherwise it uses the file-transcription operation advertised by the Gradio config.

Operations

• Transcribe one local audio file with optional diarization.
• Poll REST tasks until completion, or consume the Gradio event stream.
• Return the raw task result and an SRT transcript when the interface supplies one.

Runtime Inputs

• Reachable HTTP or HTTPS base URL supplied with --url or WHISPER_WEBUI_URL.
• Audio file path and output JSON path.
• Optional Basic Auth supplied through WHISPER_WEBUI_USERNAME and WHISPER_WEBUI_PASSWORD.
• Optional HF_TOKEN for the diarization model; never persist it in output.
• Optional model, language, diarization device, background music removal, and TLS verification settings.

Outputs

The client writes JSON containing the selected adapter and the completed API result. The LPC workflow in python/lpc_whisper_analysis.py sanitizes this response and builds the repository-specific analysis bundle.

Notes

• Prefer the REST API when both interfaces are exposed because it returns structured segments and durable task status.
• The Gradio adapter discovers component defaults at runtime so input order can change without hardcoded host settings.
• Pass remove_background_music=True to enable Whisper-WebUI's UVR filter before transcription.
• Regenerate the client if Whisper-WebUI returns a 404/405 for its documented operations or a 400/422 caused by a changed request schema.
• On Intel macOS, the verified local stack uses PyTorch/TorchAudio 2.2.2, NumPy 1.26.4, SciPy 1.11.4, Gradio 5.29.0, and huggingface-hub 0.28.1. Guard every torch.xpu.is_available() call with hasattr(torch, "xpu") because this PyTorch build does not expose the XPU namespace.