Overview

Hive's AI-Generated Music Detection API takes a video or audio clip and determines whether or not the music in the clip is AI-generated. Confidence scores are provided for each classification for easy interpretation of results. This API serves multiple purposes in the music industry, including but not limited to: ensuring that music is attributed to its rightful creators under copyright law, preventing potential copyright disputes resulting from undetected AI music, and more.

Response

The AI-Generated Music Detection API has two models and three heads. Each head contains a set of classes. Each class within a head has a confidence score ranging from 0.0 to 1.0, indicating how certain the model is of its prediction for said class. The confidence scores for each model head sum to 1.

1. AI-Generated Music: Determines if a piece of music (not including vocals) is AI-generated. This model. contains two heads.
   1. AI-Generated Music (Head): Determines if a piece of music (not including vocals) is AI-generated. This head contains two classes:ai_generated_music and not_ai_generated_music.
   2. Attribution: Determines which generative AI model created the audio clip. This head contains the following classes: udio, mubert, musicgen, riffusion, stable_audio, and suno.
2. AI-Generated Music Cover: Determines if the vocals in a song are AI-generated. Capable of detecting both deepfake audio covers of existing songs and AI-generated original songs. This head contains two classes: ai_generated_music_cover and not_ai_generated_music_cover.

When a query is made to the API endpoint, we run both models, which make classifications for each 10-second audio chunk. Their combined output is returned as a JSON response, formatted as an array of 10-second chunks with associated timestamps and classifications.

To see a complete API response object, refer to the API reference page.

Supported File Types

Video Formats:
mp4
webm
avi
mkv
wmv
mov

Audio Formats:
flac
mp3
ogg
wav
m4a