Themoor/Ai-DJ-Mixer

AI DJ Software: ML Model Reservoir 🎧

This repository hosts the official machine learning models for the AI-DJ-Software project. These models power the application's core capabilities, including automated mixing, harmonic sequencing, and genre-specific transitions.

🚀 Model Overview

Everything here is optimized for offline, low-latency performance using ONNX Runtime (int8 quantization).

1. General Genre & Audio Analysis

• Genre CNN (v1.2.0): Dual-path CNN for classifying 10 base genres.
• Mood HuBERT (v1.0.3): HuBERT-distilled classifier for semantic mood detection (6 classes).
• CLAP Audio Encoder: Generates 512-dimensional embeddings for similarity search.
• Beat Detector (v1.0.0): TCN-based onset tracking and beatgrid estimation.

2. Subgenre Specialists (Proprietary Training)

We provide fine-tuned specialist models for nuanced subgenre classification, critical for high-energy sets:

• Brazilian Funk Specialist: Detecting Carioca, Mandelão, and Automotivo.
• Phonk Specialist: Splitting American and Brazilian Phonk.
• Bass & Trap Specialist: Sub-bass energy detection.
• Techno, Trance, House Specialists: High-accuracy BPM and structure awareness.

📦 How to Use

These models are designed to be consumed by the AI DJ Desktop Application.

1. Manual Download: Place the .onnx and .onnx.data (plus .json label maps) into your local ml-sidecar/models/ directory.
2. CLI Usage:

   # Use the Hugging Face CLI to pull the full repo
   hf download Themoor/Ai-DJ-Mixer . --local-dir path/to/ai-dj/ml-sidecar/models
   

🛠 Model Technical Details

Model	Format	Precision	Size	Purpose
Specialists	ONNX	Int8	~345MB	Granular Subgenre Logic
Genre CNN	ONNX	Int8	12.4MB	Global Genre Tags
CLAP	ONNX	FP32/Int8	142MB	Similarity Embeddings
HuBERT Mood	ONNX	Int8	89MB	Mood-based Sequencing

📜 License & Credits

• Software: MIT License.
• Models: Released under Open Source for the DJ community.
• Training: Powered by PyTorch and fine-tuned on custom datasets.

Developed by Moorish.dev.