| license: apache-2.0 | |
| base_model: laion/larger_clap_music | |
| tags: [audio, coreml, clap, music-similarity] | |
| # PrepDeck — "sounds like" model | |
| The audio encoder of [LAION's CLAP](https://huggingface.co/laion/larger_clap_music) | |
| (`larger_clap_music`, Apache-2.0), converted to Core ML (half precision) for the | |
| PrepDeck macOS app. Input: CLAP's log-mel spectrogram of 10 s of 48 kHz mono audio, | |
| `[1, 1, 1001, 64]`. Output: a 512-number embedding, L2-normalised. | |
| Unchanged weights; the only edit is the bicubic resize of the spectrogram, replaced by | |
| the equivalent fixed matrix so Core ML can run it. Licence: Apache-2.0, as the original. | |