PitcheeCore

Voice analysis at the core.

A standalone C++17 voice analysis library. Through its C API it provides pitch, spectrum, model inference and scoring, with no dependency on any UI framework. Pitchee's iOS and Android apps are both built on it.

Related links
typedef struct pitchee_f0_frame_t {
    double timestamp_seconds;
    float f0_hz;
    float confidence;
    int32_t voiced;
    int32_t reserved;
} pitchee_f0_frame_t;

Feed audio in, get analysis out

The real-time F0 and spectrum interfaces take 16 kHz mono Float32 PCM. The caller handles microphone capture and resampling; Core returns analysis frames through callbacks. The spectrum interface also runs on its own, with no model loaded.

One result, every platform

For a complete recording, Core returns a UTF-8 JSON result with pitch statistics, timbre score, naturalness and the scores for both directions. iOS, Android and desktop share the same analysis interface — the same recording yields the same result, presented by each platform's own UI.

How the score is computed

A recording first passes voice activity detection to isolate actual speech; pitch tracking and acoustic embeddings follow; the score and naturalness are then computed for the feminization or masculinization direction. The formulas and parameters live in the public repository and documentation, so any result can be recomputed and checked.

Run it in your app

Models load from a local directory and run through ONNX Runtime. Core never downloads models over the network. The build guide covers platform dependencies and configuration; the API documentation specifies input requirements, threading constraints and resource ownership.