MANIFESTRO DOCS
Console
Models

Kezek

Conversation separation: one mono recording in, one track per speaker out.

Kezek decides who is who once for the whole recording, then a separator carries out that decision. The two tracks keep the original length and timing, and they always add up to the input.

Cleanup modes

ModeWhat it doesPrice per minute
noneSeparation only$0.010
cleanSeparation and noise removal$0.015
studioSeparation and studio-quality restoration$0.025

Input

  • Formats: WAV, FLAC, MP3, M4A, OGG.
  • Up to 4 hours and 500 MB per file.
  • Stereo input is mixed down to mono before processing.

Output

  • mono_pair — two mono files, one per speaker.
  • stereo — one file, speaker A on the left channel and B on the right.
  • WAV, FLAC or MP3 at 16, 24, 44.1 or 48 kHz.

Limitations

  • Exactly two speakers.
  • Quiet replies under a much louder voice may stay on the louder speaker's track.
Studio cleanup, stereo output
# 1. Register the file: the answer has its id and a pre-signed upload URL
curl https://api.manifestro.io/v1/files \
  -H "Authorization: Bearer $MANIFESTRO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"filename": "conversation.wav"}'

# 2. Upload the bytes to upload.url with the headers from upload.headers
curl -X PUT "$UPLOAD_URL" -H "Content-Type: $CONTENT_TYPE" --data-binary @conversation.wav

# 3. Create the job
curl https://api.manifestro.io/v1/jobs \
  -H "Authorization: Bearer $MANIFESTRO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "kezek-1", "file_id": "$FILE_ID", "cleanup": "studio", "output_format": "flac", "output": "stereo", "sample_rate": 48000}'

# 4. Poll until status is "completed", then download result.tracks[].url
curl https://api.manifestro.io/v1/jobs/$JOB_ID \
  -H "Authorization: Bearer $MANIFESTRO_API_KEY"