Find melodic clues in Reference Audio

Prepare 16 seconds of Reference Audio and get ABC and MIDI. Check the score against the original, finding phrases to study and differences that need attention.

1. Choose material you can hear clearly

Choose your own short instrumental fragment or licensed material. A clear lead melody with a less crowded background can make inspection easier. This is input-selection advice, not an accuracy guarantee.

The current real path verifies and accepts only exactly 16 seconds of PCM16 WAV: mono at 24 kHz or stereo at 48 kHz. Export one supported format with your existing audio tool and keep the original. Note one clear pitch movement or rhythm in the source for later comparison.

The 64 MiB / 600-second upload budget is a storage limit, not transcription support for those lengths. This operation does not recognize lyrics or automatically create a Cover.

2. Upload to the Project, then transcribe once

Continue in the Morning song Project and upload the file as Reference Audio. Open “Reference transcription” in the formal Web, select the original file, check readiness, then choose “Start transcription”. When completed, inspect the score and download ABC/MIDI. The original Reference Audio stays unchanged. Follow the workspace transcription guide for complete steps. Expand the supplement below when you need API operations; the documentation page uploads no files.

Current entrypoint: the complete local Swagger path

1. Open the running application API's http://127.0.0.1:8000/docs. Use a ready real Runtime for model work; fake mode returns identified test results.
2. Reuse your Project id. If starting here, send {"name":"Morning song"} to POST /projects and record the returned id.
3. In POST /projects/{project_id}/assets, enter that Project id and choose the WAV in file. Submit and record the returned Asset id.
4. Use the JSON below in POST /projects/{project_id}/transcriptions, replacing the placeholder. Submit once and record the returned Job id.
5. Read the same Job with GET /projects/{project_id}/jobs/{job_id} until completed or explicitly failed. A completed result includes score_id, abc_asset_id, and midi_asset_id.
6. Read score information with GET /projects/{project_id}/scores/{score_id}, then download ABC and MIDI separately through GET /projects/{project_id}/assets/{asset_id}/content. Keep them in your Morning song folder.

{
  "reference_asset_id": "<REFERENCE_ASSET_ID>"
}

3. Compare by listening, not just file existence

1. Play the original Reference Audio and listen again to the phrase you noted.
2. Read the ABC text or use your own ABC reader to inspect bars, meter, and pitch. Open MIDI in your existing MIDI player or music software.
3. Compare melodic movement and rhythm with the original. Note where a phrase differs. A downloadable score is not necessarily an exact transcription.

The application validates complete files and imports the Score; that does not establish accuracy for arbitrary music. The integrated editor can correct notes, audition, save and regenerate in the formal Web. Continue with melody/full Cover while retaining the inspectable Reference origin.

If the format is rejected, fix export settings

422 reference_profile_unsupported means the audio profile is outside the current range. Check exactly 16 seconds, 16-bit integer PCM WAV, and supported channels and sample rate. Renaming an extension cannot change encoding. Export a supported file and upload it as new Reference Audio.

If submission acknowledgement is uncertain or waiting disconnects, query the known Job first. Repeated uploads or submissions are not recovery. Download the same completed result, or retain an explicitly failed error and follow its recovery guidance.

Next idea: use the reference as a listening question

Note one feature of the fragment: a sparser rhythm, a smoother melody, or the contour of a phrase. Inspect, save and select this Score, then use melody/full Cover to change only style and hear another accompaniment.

Try changes to style and lyrics →

Source version: 1ce3268Page source: docs/learn/reference.en.md