audio-ml

A desktop app to fine-tune, generate and perform live music with diffusion models. Runs on MacOS, Windows and Linux.

Code to train a custom time-domain autoencoder to dereverb audio

Dual-model speech AI toolkit for speaker verification and speaker-aware diarization, with streaming inference, meeting analysis, long-audio monitoring, and speaker-bank integration.

Local MCP server + CLI turning YouTube & local audio into rich sonic signatures. Extracts BPM, section-by-section key, vocal presence, transient punch, and 512-dim CLAP vibe embeddings. Powered by Demucs stem separation & librosa. 100% private, offline-first, and GPU-accelerated with graceful CPU/HPSS degradation.

Unified dual-teacher distillation (ReDimNet + ASSIST) into Wav2Vec2 for speaker verification and deepfake detection

From-scratch audio-AI research system + 99-lesson bilingual course — DSP · ASR · Music · LLM/RAG, every core hand-written, no API wrappers.

An intelligent, automated video switcher for live performances

This repository consists of projects where GANs were used to work with audio data like music generator etc.

Key Features: Simple VAE architecture with encoder/decoder Synthetic music data generation for training Interactive training with progress tracking Music generation from latent space sampling Audio conversion and playback Downloadable audio files

Real-time speech enhancement pipeline — custom-trained U-Net denoising model, ONNX inference, Overlap-Add synthesis, and virtual audio routing for Teams, Zoom, and DAW use. CPU-only, no cloud dependency.

Stable Audio LoRA Trainer of salty goodness

Engine identification using acoustic signal analysis and machine learning to classify 8 vehicle types. Audio signals are processed using FFT and feature extraction, and a multi-class model predicts vehicle categories based on their unique sound patterns.

Audio analysis in javascript/typescript

Edge-deployable keyword spotter: INT8-quantized DS-CNN on Google Speech Commands, exported to ONNX, with fp32 vs INT8 benchmarks, a live mic demo, and a C++ inference harness.

Convert Meta's HTDemucs (Hybrid Transformer Demucs) to Apple Core ML. Real-valued STFT/ISTFT wrapper, manual MHA decomposition, pre-computed overlap-add. Includes Swift example.

Privacy-first baby cry monitor. Runs fully on-device: local cry detection, night logs, and morning digests.

Audio file processing pipeline with GPT-4-powered error diagnosis — detects codec issues, sample rate mismatches, and corruption artefacts with automated remediation suggestions.