Archer VCS

About this project

Archer is a real-time vocal collaboration system designed to listen to a singer and respond as a musical partner rather than simply process audio after the fact. The live pipeline captures microphone input, tracks pitch with YIN or RMVPE, analyzes rhythm and phrasing, and uses a configurable harmony engine to decide how the system should respond. Responses can take forms such as unison, thirds, fifths, octaves, drones, and call-and-response patterns.

The selected response is rendered through a DSP-based vocable synthesizer using techniques such as sinusoidal, wavetable, and formant synthesis. An optional RVC voice-conversion sidecar adds neural singing voice generation while keeping the heavier PyTorch workload separate from the real-time audio loop.

Alongside the musical pipeline, the project includes a FastAPI research application with Cree morphological analysis, Cree-to-English translation, sentiment analysis, faster-whisper transcription, Vosk live captions, pitch analysis, and a notes-grounded assistant. The architecture is designed so individual model-dependent components can fail without taking down the rest of the interactive system.