| --- |
| license: mit |
| tags: |
| - audio |
| - stem-separation |
| - audio-transformation |
| - deep-learning |
| --- |
| # 🔱 Audio Reborn |
| A professional GPU-accelerated audio transformation pipeline designed to bypass Content ID fingerprints using advanced AI synthesis. |
|
|
| ## 🏗️ Project Architecture (GPU-Only) |
|
|
| ### Stage 1: Deep Audio Analysis (`1_Analysis`) |
| - **Tools:** Librosa (GPU-backed via CuPy/Torch), Crepe. |
| - **Goal:** Extract BPM, Key, Pitch, and Spectral Centroid. |
| |
| ### Stage 2: AI Stem Separation (`2_Stem_Separation`) |
| - **Tools:** Demucs v4 (HTDemucs) - CUDA Enabled. |
| - **Goal:** Separate Vocals, Drums, Bass, and Other. |
| |
| ### Stage 3: Multi-Layer Transformation (`3_Transformation`) |
| - **Tools:** Pedalboard (VST3/GPU), Audiomentations (Torch-based). |
| - **Goal:** Phase shifting, Micro-latency, Random EQ, Reverb. |
|
|
| ### Stage 4: AI Identity Regeneration (`4_Regeneration`) |
| - **Tools:** RVC v2 (CUDA), Meta AudioCraft (MusicGen/AudioGen). |
| - **Goal:** Complete Timbre replacement and Melody-conditioned synthesis. |
| |
| ### Stage 5: AI Mastering & Quality Check (`5_Mastering`) |
| - **Tools:** Matchering, Pyloudnorm, VISQOL. |
| - **Goal:** Professional loudness and artifact verification. |
|
|
| --- |
| ## 🛠️ Infrastructure Requirements |
| - **OS:** Windows (PowerShell) |
| - **GPU:** NVIDIA (CUDA Toolkit 11.8/12.1) |
| - **Python:** 3.10+ (Inside `venv`) |
|
|
