AI marks in Midsummerr audio

Every audio file Midsummerr delivers carries machine-readable marks showing it was made with AI:

• An inaudible watermarkin the sound itself (Meta's open-source AudioSeal). In our tests it survives MP3 and AAC re-encoding, volume changes and EQ; heavy time-stretching can remove it.
• Content Credentials (C2PA): signed, time-stamped metadata naming Midsummerr as the system that made the audio. Re-encoding the file removes it; that is normal.
• A tag in MP3 files (ID3 TXXX:AI_GENERATED).

When a file combines human-recorded audio with AI-generated voices, music or sound effects, the marks say the file contains AI-generated elements and the watermark covers the AI parts where we can locate them. Previews played inside the studio are not marked; files you download are.

Check a file for free below or with POST https://api.midsummerr.com/api/v1/provenance/detect. We use the file only to look for marks and delete it immediately. You get a signed receipt (file hash, detector version, time). “No mark found” means no Midsummerr mark was found, not that the audio is human-made.

Our signing certificate fingerprint: E5:0C:CD:69:3A:81:32:ED:15:9A:4D:53:AD:2C:9E:A8:56:BF:98:01:D8:E9:61:70:81:51:0D:1F:45:6D:8F:DA. Questions: legal@midsummerr.com.