A dual-domain signal equalizer with AI-powered source separation, built with Flask and a dark-themed web UI. Supports ECG biomedical signals, music, voice, and animal audio — processed via FFT, wavelet transforms, and deep learning models.
- 5 Signal Modes — ECG, Music, Voices, Animals, and Generic
- Dual-Domain Equalization — FFT-based filtering and wavelet denoising (mode-optimized wavelets)
- AI Denoising & Source Separation
- ECG: Fully Convolutional Denoising Autoencoder (FCN-DAE) via TensorFlow/Keras
- Music: Stem separation (drums, bass, vocals, guitar, piano, other) via Demucs
- Voices: Male/female speaker separation via SpeechBrain SepFormer
- Animals: Dog, bird, cat, frog isolation via AudioSep with text-query inference
- Interactive Visualizations — Time-domain waveforms, FFT spectrum, spectrogram, and CWT scalogram
- Stem Mixer — Per-stem gain sliders for music, voice, and animal modes
- Audio Playback — Listen to original, filtered, and separated audio directly in the browser
- Settings Persistence — Save and load equalizer presets as JSON
| Layer | Technology |
|---|---|
| Backend | Python, Flask |
| Signal Processing | NumPy, SciPy, PyWavelets |
| AI — ECG | TensorFlow / Keras |
| AI — Music | Demucs (PyTorch) |
| AI — Voices | SpeechBrain, Torchaudio, Librosa |
| AI — Animals | AudioSep (PyTorch) |
| Frontend | Vanilla JS, Chart.js |
signal-equalizer/
├── app.py # Flask backend — all API routes and signal processing logic
├── index.html # Single-page frontend UI
├── settings/ # Saved equalizer presets (auto-created)
└── AudioSep/ # AudioSep repo (external, see setup below)
- Python 3.9+
ffmpeg(required for MP3 support via pydub/torchaudio)
# Clone the repository
git clone https://github.com/your-username/signal-equalizer.git
cd signal-equalizer
# Install core dependencies
pip install flask numpy scipy pywavelets
# Optional — install AI backends as needed (see below)
⚠️ Note: Pre-trained model weights are not included in this repository due to their large file sizes. You will need to download each model separately following the instructions below.
Each AI feature requires its own set of dependencies. Install only what you need:
ECG Denoising (TensorFlow)
pip install tensorflowMusic Separation (Demucs)
pip install demucs torchaudio librosaVoice Separation (SpeechBrain)
pip install speechbrain torchaudio librosaAnimal Sound Separation (AudioSep)
# Clone the AudioSep repo
git clone https://github.com/Audio-AGI/AudioSep
# Update AUDIOSEP_REPO_DIR in app.py to point to your local clone
# Download the checkpoint (1.26 GB) from HuggingFace:
# https://huggingface.co/Audio-AGI/AudioSep
# Place it at: AudioSep/checkpoint/audiosep_base_4M_steps.ckpt
pip install torch torchaudiopython app.pyThen open http://localhost:5000 in your browser.
| Endpoint | Method | Description |
|---|---|---|
/api/upload |
POST | Upload a signal file (CSV, DAT, WAV, MP3) |
/api/fft_equalize |
POST | Apply FFT-based equalization |
/api/wavelet_equalize |
POST | Apply wavelet denoising |
/api/ai_denoise |
POST | Run AI denoising / source separation |
/api/mix_voice_stems |
POST | Adjust male/female voice stem gains |
/api/mix_music_stems |
POST | Adjust music stem gains |
/api/mix_animal_stems |
POST | Adjust animal stem gains |
/api/spectrogram |
GET | Retrieve spectrogram data |
/api/scalogram |
GET | Retrieve CWT scalogram data |
/api/audio |
GET | Stream processed audio as WAV |
/api/settings/save |
POST | Save current settings preset |
/api/settings/load |
POST | Load a settings preset from file |
/api/settings/default |
GET | Load the default ECG preset |
| Mode | Wavelet | AI Model | Stems |
|---|---|---|---|
| ECG | sym5 |
FCN-DAE (TF/Keras) | — |
| Music | db4 |
Demucs | drums, bass, vocals, guitar, piano, other |
| Voices | haar |
SepFormer (SpeechBrain) | male, female |
| Animals | coif3 |
AudioSep | dog, bird, cat, frog, other |
| Generic | db4 |
FCN-DAE (TF/Keras) | — |
.csv— comma-separated signal values.dat— whitespace-separated signal values.wav— PCM audio.mp3— compressed audio (requires ffmpeg)
This project is open source. See LICENSE for details.