# Separate Overlapping Voices

AudioPod separates overlapping voices with its AudioDiarize engine: where two people talk at once, it isolates each voice and patches it into that speaker's own track, so crosstalk is not left in both. It runs on every plan, including free Basic, at 330 credits per minute, with no extra charge for overlaps.

Canonical URL: https://audiopod.ai/speaker-separation/overlapping-voices
Last updated: 2026-10-16

## Facts

| Fact | Detail |
| --- | --- |
| Default behaviour | Crosstalk is separated: each speaker's track carries only their own voice |
| Shortest overlap handled | 0.25 s when the crosstalk detector confirms it; 0.4 s for a plain overlap of two speaker turns |
| How voices are matched | A voice profile per speaker, built from up to 60 seconds of that speaker talking alone |
| When it is not sure | It keeps the original mixed audio for that stretch rather than guess which voice is whose |
| Overlap modes (API) | overlap_mode = "separate", "duplicate" or "exclusive" |
| Singers or several voices at once | POST /api/v1/speaker/separate: 2–4 simultaneous voices, with an option to isolate vocals from music first |
| Output | One full-length track per speaker, as WAV and MP3 |
| Price | 330 credits per minute of audio, the same with or without overlaps |

## Three ways to handle crosstalk

The same overlap, in a two-person recording, under each mode.

| Mode | Speaker A's track | Speaker B's track | Use it when |
| --- | --- | --- | --- |
| separate (default) | Only A's voice | Only B's voice | You will edit, level or transcribe each person on their own |
| duplicate | The mixed audio, both voices | The mixed audio, both voices | You want nothing processed: every track keeps exactly what was said over it |
| exclusive | The mixed audio, if A is the dominant speaker | Silent there | You want each moment in exactly one track, as in a cut-down edit |

## How to do it

1. Open Speaker Separation and upload the recording or paste a link.
2. Leave the speaker count on Auto detect, or set it if you know it.
3. Check the quote (330 credits per minute) and start the job.
4. Play each speaker's track; overlaps are already split between them.
5. Over the API, pick a different overlap_mode on POST /api/v1/speaker/extract.

## When the mix is kept instead

Separating an overlap means deciding which voice belongs to whom. Where that cannot be done reliably, the stretch keeps the original audio:

- A speaker who never talks alone has no voice profile to match against.
- A separated voice does not match any speaker in the recording.
- The separated voices do not add back up to the original sound.
- The overlap is too quiet to judge against the rest of the speech.

## What it cannot fix

- Interjections shorter than 0.25 seconds stay as they were recorded.
- A speaker with less than 0.5 seconds of speech in the whole file gets no track of their own.
- Heavy room echo smears voices together; a close microphone per person separates far more cleanly.

## FAQ

**Can AudioPod separate two people talking at the same time?**
Yes. Where two speakers overlap, AudioPod isolates each voice and puts it in that speaker's own track. Each speaker is recognised from up to 60 seconds of their clean speech elsewhere in the file. Where it cannot tell the voices apart reliably, it keeps the original mixed audio for that stretch instead of guessing.

**Does separating crosstalk cost more?**
No. Speaker separation is 330 credits per minute of audio however much overlap the recording has. The free Basic plan's 5,000 monthly credits cover about 15.1 minutes. See /pricing for plans.

**What are the separate, duplicate and exclusive modes?**
They decide what happens to an overlap. Separate gives each speaker only their own voice. Duplicate leaves the mixed audio in every track that was talking. Exclusive keeps the mixed audio only in the dominant speaker's track and silences it in the others. Separate is the default; the API's overlap_mode parameter picks another.

**Can it split singers or a choir singing together?**
For simultaneous voices that never take turns, use POST /api/v1/speaker/separate: it splits 2 to 4 overlapping voices and can isolate the vocals from the music first. For lead versus backing vocals in a song, the Stem Splitter's vocal splits are the better fit.

**Why does a short overlap still have both voices in it?**
Overlaps under 0.25 seconds are left alone, and longer ones keep the mixed audio when the separation cannot be trusted — for example when one of the speakers never talks alone anywhere in the file.

## Related

- [Speaker Separation](https://audiopod.ai/features/speaker-separation) — The tool, the free demo and the speaker guides.
- [Long recordings](https://audiopod.ai/speaker-separation/long-recordings) — Length and upload limits per plan, batches and video files.
- [Speaker-labeled transcript](https://audiopod.ai/speaker-separation/speaker-labeled-transcript) — Who said what, as text, next to the tracks.
- [Male and female vocal split](https://audiopod.ai/stems/vocals-and-choir) — For singers in a song rather than speakers in a recording.

## Pricing

Plan prices: [pricing](https://audiopod.ai/pricing). Start: [Separate a recording](https://audiopod.ai/auth/signup?callbackUrl=%2Fdashboard%2Ftools%2Fspeaker-separation).