Credits

What RAAYDR is built with.

Moods

RAAYDR listens to every track to suggest the moods it sits in. It does that with CLAP, a model that learned to match sound to words.

CLAP: Learning Audio Concepts From Natural Language Supervision (pretrained model, 2023 weights), by Benjamin Elizalde, Soham Deshmukh, Mahmoud Al Ismail and Huaming Wang, Microsoft. Published at zenodo.org/records/8378278 under the Creative Commons Attribution 3.0 United States licence. We use the weights as published, without changes. The code that runs them, microsoft/CLAP, is MIT licensed.

It also reads the words. Where an artist has not written their lyrics down, it listens for them with Whisper, by OpenAI, published under the MIT licence, run with faster-whisper by SYSTRAN, also MIT licensed. What it hears is a draft for us, not something we show: an artist checks their words before anything is displayed.

What the words are about is read with paraphrase-multilingual-MiniLM-L12-v2 from Sentence Transformers, by Nils Reimers and Iryna Gurevych, under the Apache 2.0 licence.

Tempo and loudness are measured with librosa, ISC licensed.

The model suggests. A person at RAAYDR checks how well it hears before anything it suggests goes live, and an artist can always ask us to move their track.