> ## Documentation Index
> Fetch the complete documentation index at: https://docs.deploy.com/llms.txt
> Use this file to discover all available pages before exploring further.

# AI-Generated Music Detection

> Detect AI-generated music, identify generative sources, and recognize AI-generated vocals.

Deploy's AI-Generated Music Detection API analyzes audio and video content for signs of AI-generated music and AI-generated vocals. A single endpoint combines AI-generated music detection, generative source attribution, and AI-generated vocal detection.

The API processes media in 3-second audio chunks and returns classification confidence scores with associated timestamps, helping identify AI-generated content throughout an audio or video clip.

## Detection Capabilities

> [card] AI-Generated Music Detection
>
> Determine whether a piece of music is AI-generated and identify its likely generative model or platform.

> [card] AI-Generated Vocal Detection
>
> Identify AI-generated or manipulated vocals, including deepfake audio covers of existing songs and AI-generated original vocals.

## AI-Generated Music Detection

The AI-Generated Music model determines whether a piece of music was generated using AI. It is trained to distinguish AI-generated music from human-created music.

The model contains two classification heads:

* **Generation Classification:** Determines whether the music is AI-generated.
* **Source Attribution:** Identifies the likely generative AI model or platform responsible for creating the music.

### Classification Outputs

**Generation Classification**

| Classification | Description |
| - | - |
| `ai_generated_music` | Confidence that the music is AI-generated |
| `not_ai_generated_music` | Confidence that the music is not AI-generated |

**Source Attribution**

The source attribution head identifies the likely generative model or platform.

Supported source classes include:

* `duobao`
* `google`
* `heartmula`
* `loudly`
* `minimax`
* `mubert`
* `mureka`
* `musicgen`
* `riffusion`
* `stable_audio`
* `suno`
* `udio`
* `yue`
* `ace_step`
* `sonauto`
* `elevenlabs_music`
* `rvc`

Each classification returns a confidence score between 0 and 1. The confidence scores within each classification head sum to 1.

> [!NOTE]
>
> Source attribution is separate from AI-generation detection. Music may be classified as AI-generated even when the specific generative source cannot be identified.

## AI-Generated Vocal Detection

The AI-Generated Music Vocal model identifies vocals that may have been generated or manipulated using AI.

It can detect:

* **AI-generated music covers:** AI-generated vocals that imitate the voice of an existing artist or singer to create a cover of an existing song.
* **AI-generated original vocals:** AI-generated vocals used in original songs that were not created as covers of existing music.

### Classification Outputs

| Classification | Description |
| - | - |
| `ai_generated_music_vocal` | Confidence that the vocals are AI-generated or AI-manipulated |
| `not_ai_generated_music_vocal` | Confidence that the vocals are not AI-generated or AI-manipulated |

Each classification returns a confidence score between 0 and 1. The scores within this classification head sum to 1.

## C2PA Metadata

When available, the API also returns C2PA metadata describing the media's origin, such as information about its generative source.

C2PA metadata can provide additional context about how and where media was created.

> [!NOTE]
>
> C2PA metadata may be removed or falsified. We recommend interpreting it alongside the model's detection results rather than relying on metadata alone.

## Detection Results and Response Structure

The API runs both detection models on the submitted media. Each model analyzes the audio in 3-second chunks and returns classification results for each chunk.

The combined response is returned as an array of timestamped 3-second chunks, with classification results and associated confidence scores for each detection head.

## Recommended Detection Thresholds

We recommend the following thresholds as starting points for production workflows.

| Detection Type | Recommended Threshold |
| - | - |
| AI-Generated Music | Median score ≥ 0.99 on the full clip |
| AI-Generated Vocals | Median score ≥ 0.995 on the full clip |

> [!NOTE]
>
> These thresholds are general recommendations. For optimal performance, calibrate them using representative data from your use case and your tolerance for false positives and false negatives.

## Supported File Formats

> [tab] Video
>
> * MP4
> * WEBM
> * AVI
> * MKV
> * WMV
> * MOV

> [tab] Audio
>
> * FLAC
> * MP3
> * OGG
> * WAV
> * M4A

## API Reference

> [!NOTE]
>
> This API is currently only available on our legacy API. Please refer to our [AI-Generated Music Detection API Reference](https://docs.thehive.ai/reference/ai-generated-music-detection-1) for more details.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.