Description & Run

This tab is the “control room” for separation runs.

Description

Enter a short natural-language description of what you want to extract.
Examples: “speech”, “engine noise”, “birds”, “applause”.

Tips:

• Keep it short and concrete.
• If results are inconsistent, add anchors and reduce chunk length.

Chunking

Chunking settings control how long audio is split for inference.

• Max chunk length (s): window size
• Overlap (s): overlap between adjacent chunks (helps continuity)

See Chunking for recommended starting points.

Runtime settings

Choose compute/runtime settings:

• Device: auto, cuda, or cpu
• Half precision: FP16 on supported GPUs
• Predict spans: enable span prediction output (if supported)
• Reranking: number of candidates considered
• No resample: keep native sample rate (if supported)

See Runtime Settings for details.

Run / Abort and progress

• Run starts a backend call.
• Abort cancels the current run.

Progress display:

• A progress bar
• Percentage (centered under the bar)
• Current phase text (centered under the percentage, wrapped if needed)

Parameters

The Parameters section summarizes the backend call configuration and provides:

• Copy snippet: copies a reproducible Python snippet to the clipboard
• Save snippet .py…: writes the snippet to disk

See Backend API for how the snippet maps to patchbay_backend.Config and run_pipeline.

UI note: The Parameters table is scrollable so long configs remain usable.

Last updated: 2026-01-24