Hatsumi

Real-time voice changer for macOS, built in Rust. Captures mic input, processes it through a DSP pipeline (Rubber Band pitch/formant shift + effect chain), and routes output to a virtual audio device so any app (Zoom, Discord, etc.) can use it as a mic. Controlled live from a ratatui TUI.

Architecture

graph TD
    MIC[Mic / CoreAudio]
    CAP[cpal input stream]
    RB[Ring buffer]
    GATE[Noise gate]
    SHIFT[Rubber Band pitch/formant shift]
    FX[Effect chain: EQ / ring mod / clip / reverb]
    OUT[cpal output stream]
    BH[BlackHole]

    MIC --> CAP --> RB --> GATE --> SHIFT --> FX --> OUT --> BH

    BH --> Zoom
    BH --> Discord
    BH --> Other[any app]

Loading

Code layout: src/lib.rs is the core library (audio device handling, dsp pipeline, effects, engine, presets); src/main.rs + src/ui.rs are the CLI/TUI binary.

Presets

Built-in: Woman, Villain, Robot, Kid, Radio.

User presets are read from ~/.config/hatsumi/presets.toml (or --presets FILE); a preset with a built-in's name replaces it. Example:

[[presets]]
name = "Ghost"
description = "Low haunted whisper"
semitones = -9.0
formant_scale = 0.9

[[presets.effects]]
reverb = { mix = 0.4, decay = 0.85 }

[[presets.effects]]
low_pass = { freq_hz = 4000.0 }

Available effects: low_shelf / high_shelf (freq_hz, gain_db), high_pass / low_pass (freq_hz), ring_mod (freq_hz, mix), soft_clip (drive), reverb (mix, decay).

Status

See ROADMAP.md for what's next.

Requirements

• macOS (CoreAudio)
• Rust toolchain
• BlackHole virtual audio driver (create an aggregate/virtual device named "Hatsumi", or pass --output)

Setup

make setup   # register git hooks

Run

make run                              # default preset (Woman), TUI
cargo run -- robot                    # start on a preset
cargo run -- --list-devices           # show input/output devices
cargo run -- ghost --presets my.toml  # custom preset file
cargo run -- --input realme --output blackhole

TUI keys

Key	Action
↑/↓ or j/k, Enter	select / apply preset
1–9	apply preset directly
+/-	pitch ±0.5 semitone
[/]	formant scale ±0.05
g, ,/.	toggle noise gate, threshold ±3 dB
r	reset tuning to preset defaults
q/Esc	quit

The status panel shows live pitch/formant values, over/underrun counters, and a continuously measured latency estimate; the level meters show input and processed output in dBFS.