demorphus
For an AI song with vocals, upload the complete mix to Cleaner and retain the processed export with its SH Labs report. Fingerprint processing and vocal repair answer different questions: a lower AI score does not establish that distorted words or a metallic voice have been repaired.
Full mix or vocal stem?
Use the complete song when that is the file you plan to release. A report on a vocal stem does not describe the complete mix, and a report on the complete mix does not assign a score to each singer or instrument.
If you are repairing a specific syllable, start with the vocal-artifact guide. Stem repair may belong earlier in production. Once the mix is final, use that export for processing and analysis.
Check the voice where changes are easiest to hear
- Exposed verse: intelligibility, breath and the identity of the voice.
- Chorus: lead separation from harmonies and dense backing.
- Sustained note: tone, pitch stability and unnatural modulation.
- Consonants and endings: sharp syllables, clipped words and reverb tails.
Compare at similar playback levels without modifying the files submitted for detection. Record timestamps and concrete observations. This is a quality review, not a requirement to find an audible improvement: the intended song should remain usable.
Separate transposition from vocal damage
A whole-mix pitch change can alter the song’s key and vocal register. A voice sounding lower is a different observation from a word becoming unintelligible. Record both separately, particularly if the song must match live accompaniment.
The pitch-shifting guide explains how to test that variable. The quality guide covers level, frequency detail and listening records.
Use the returned classification accurately
SH Labs describes detection of generated and processed/hybrid material. Its result is a classification of the submitted audio, not proof of the singer’s recording history. For a real vocal over an AI backing, use the separate hybrid-song guide.
No vocal-only success rate is published here. The internal results combine different tracks; this page does not infer a vocal subgroup result from that total.
Cleaner selects the output automatically. Download its audio and report, and use all plans to choose access for your release schedule.
Prepared with AI assistance. Sources below.