Spectral fusion

Compare and spectrally combine treatment candidates made from the same input.

Audio editor showing a real recording waveform, spectrogram and tools
A real recording in the editor. This screenshot demonstrates the workflow; evaluate treatment results using your own audio.

When should you choose this tool?

One denoise treatment may preserve word endings while another suppresses background more effectively. Fusion creates candidates from one fixed input and lets you choose which to combine. Review the individual candidates before judging the combined result.

How it changes audio

Complex spectral fusion uses agreement between comparable candidates across time and frequency. A fixed shared input is essential; unrelated recordings do not automatically become compatible candidates.

What do the parameters control?

Input and candidates
Fix the input, make targeted treatment candidates and name them distinctly. Names are your working labels.
Inclusion and weight
Include only useful candidates. More weight gives a candidate greater influence; it is not a quality score or intelligibility measurement.
Input channel; Candidate count
Choose Left or Right for stereo input; the channel is fixed after the first candidate. Up to 4 candidates are supported, and fusion needs at least 2 included candidates. Channel selection is not automatic stereo fusion.
Result blend
0–100%. Blends the starting audio with fusion. Lower values retain more original input; recalculate fusion after changing it.
Alignment limit
0–120 ms. Bounds permitted candidate time shifts. A larger limit does not automatically improve the result or make different utterances compatible.
Spectral gain limit
0.1–2×. Limits excessive gain in individual frequency components. A low cap can suppress useful sound; review higher values alongside the residual.
Alignment and spectral inspection
Time shift is shown in ms and level alignment in dB. Agreement and clamped-bin percentage describe relationships between candidates, not word correctness. A changed configuration marks the previous result separately and requires recalculation.

Try a short passage

  1. Choose a short passage you can review carefully. Make different candidates aimed at the same problem.
  2. Listen to candidates individually. Exclude versions that omit speech or follow the wrong source.
  3. Create a fusion Preview and compare it with the original and strongest individual candidate. Keep it only if the benefit is audible.

What to listen for

Fusion should preserve the word or syllable you are examining. Smoother background or stronger agreement is insufficient if the phrase loses content.

Limits of the method

Combining several flawed estimates does not establish an authentic original. Long passages or many candidates may exceed available memory.

Preview covers a short passage. After applying, review the full treated section; you can undo the step from history.

Related problems and workflows

Other useful tools