Separate overlapping voices in a conversation recording
Two estimated speech sources or reference-guided extraction: which to try and how to review it.
Someone else speaks over the important sentence. Both voices share a track, so simply lowering one affects the other. Separation may help through estimation, with the common original retained for comparison.
Do you have a clean target-voice reference?
If the target also speaks alone, select a clean reference for Extract speaker. It should contain one voice. Another voice or strong noise in the reference can misguide processing. Its name is a working label, not an identity finding.
Without a reference, compare two estimated outputs
Speech separation creates two new mono clips. Begin with a short passage and listen to both. Preview shows the first estimate, which alone cannot establish the quality of the second. Output order can swap after silence.
Check overlap and solo words too
An estimate may lose syllables or leak the competing voice during overlap. Check solo utterances before and after it too. Extraction Residual is input minus result, not automatically a clean track of the other speaker.
Mono processing changes the sound field
Both tools use speech-oriented 16 kHz mono processing. Returning output to the project sample rate does not restore original high frequencies or stereo space. Music backgrounds and more speakers make local review especially important.
Keep the estimate alongside the original
A clearer estimate can assist listening but cannot by itself establish an uncertain word. Preserve the original and processing history. Seek specialist examination when interpretation affects a consequential decision.
Try your own recording
Start with a short passage you can review carefully. Local editing and Preview are free; finished-clip export is paid. The editor’s price calculation shows the current cost.