# Overview

The [**AI Song Detector**](https://pex.com/ai-song-detector/) identifies whether an **audio file contains AI-generated music**. It provides a simple API for submitting an audio file and receiving a classification result with accompanying details.

The detector is optimized for **music**, specifically individual songs, and produces one prediction per file.

Submitted audio files, whether uploaded directly or retrieved from a provided URL, are processed and then discarded. They are not stored on our servers nor used for training.

***

### What the API Does

* Analyzes an audio file, either uploaded directly or fetched from a provided URL.
* Determines whether the musical content is AI-generated.
* Returns a JSON response containing:
  * classification result - is it AI-generated or not,
  * classification confidence score,
  * the detected AI generation platform (e.g., Suno, Udio), if known and detected with high enough confidence,
  * additional details.

The API is designed for straightforward integration into existing media workflows and high-volume use cases. It can process millions of songs per day, making it suitable for large-scale catalog scanning.

#### AI Generation Platform Detection

When the detector classifies a file as AI-generated, it may also identify the AI generation platform used to create the music (such as Suno, Udio, and others).

This information is returned only when the detector reaches a sufficiently high confidence for platform attribution. If platform confidence is not high enough, the response will still indicate that the content is AI-generated, but the platform field will be empty.

***

### Expected Input

The detector is designed to analyze music, not generic audio. 

Files **should** contain:

* A single musical track
* Minimal non-music content (speech, silence, etc.)

Files **shouldn't** contain:

* UGC compilations
* Voiceovers
* DJ mixes
* Mashups
* Audio dominated by silence or speech

***

### Supported Audio Formats & Limits

* **Supported formats**: MP3, WAV, FLAC, AAC, M4A/MP4, OGG, WEBM, and other common audio formats
* **Minimum duration**: 30 seconds
* **Maximum duration**: 15 minutes
* **Max file size**: 100 MB
* Stereo, mono, and variable bitrates are supported

Files outside these limits will return statuses such as `too_short`, `too_long`, or `not_enough_music`.

***

### Input Methods

The API supports two equivalent ways of providing audio for analysis:

* **Direct file upload** as part of the request.
* Providing a **publicly accessible URL** from which the API will retrieve the audio file.

Both methods are subject to the same format, duration, and size limits, and both produce identical classification results.

***

### How the API Works

1. **Authenticate** using OAuth 2.0 Client Credentials
2. **Submit** an audio file, either by uploading it directly or by providing a URL to the file
3. **Receive** the classification result in a JSON response

Each request accepts one file and produces one output. Batch submission is not supported.

***

### Changelog

This changelog documents notable updates to the Pex AI Song Detector API, including new features, improvements, and model attribution enhancements.

#### v4 - February 24, 2026

Improved:

* Improved model attribution accuracy.
* Increased overall detection accuracy, with the focus on reducing false positives in challenging edge cases.

#### v3 - January 28, 2026

Added:

* Introduced **two new model attribution classifiers**: **Mureka** and **Sonauto**.
* The model attribution classifier supports a defined set of platforms listed above.\
  Detection of AI-generated music is broader and not limited to these platforms. In such cases, `predicted_model` may be `null`.

#### v2 - January 7, 2026

Added:

* **Model attribution classifier** to identify which model most likely generated the queried song, when the detector has strong attribution confidence.
  * New response fields:
    * `predicted_model` (string)
    * `predicted_model_score` (float)
* Expanded attribution coverage with targeted improvements for **ElevenLabs** and **Boomy**, while maintaining broad attribution across other leading AI music generators.

Improved:

* **Higher detection accuracy**: internal evaluation shows an estimated **\~10% relative lift** versus v1 (varies by content type, genre, and audio quality).

#### v1 - December 1, 2025

Added:

* **Initial release** of the Pex AI Song Detector API.
* **Broad support** across major AI music generation platforms.
* **Per-request confidence scoring** via `ai_score`.

***

### Next Steps

Continue to [API Documentation](https://docs.pex.com/ai-song-detector/api-documentation) for:

* Authentication flow
* Detector endpoint details
* Response structure
* Error handling
* Example implementations