Separate Overlapping Speakers
When two people talk over each other on one track, this tool pulls the dominant voice forward and reduces the overlap, making cross-talk easier to follow and transcribe.
AI model This tool runs a generative AI model on a GPU, so it takes longer than the instant filters. Free to try; paid plans get priority on the GPU.
How it works
Overlapping speech is separated by source so the foreground speaker is isolated from the competing voice and room. The result is a clearer single-speaker track from a messy cross-talk recording.
What it's good for
- Cross-talk in interviews
- Single-mic two-person recordings
- Cleaning up debate audio
- Transcription prep for overlap
Details
- Engine
- Mel-Band RoFormer
- Formats
- MP3, WAV, M4A, FLAC, OGG, AAC
- Price
- Free to try
Frequently asked questions
It isolates the dominant near-field speaker and suppresses the overlap. Full per-speaker diarization into separate tracks is on our roadmap; today it cleans the foreground voice.
Yes — reducing the competing voice and room makes speech-to-text far more accurate on overlapping sections.
Recordings where your target speaker is closest to the mic separate best, since proximity gives the model a strong foreground cue.
Speaker separation is built for two voices talking over each other on a clean track, while target-speaker extraction pulls one voice out of a wider mess of voices, music and noise.
It is tuned for two overlapping voices; with three or more it still lifts the dominant near-field speaker, but the result is less clean than a true two-person cross-talk recording.
A typical interview segment processes in under a minute, scaling with clip length rather than with how much overlap there is.
Equal-volume voices are the hardest case and leave more of the competing speaker behind; the tool performs best when your target is clearly the closer, louder voice.