ComfyUI-Lyrebird

Voice cleanup and a voice changer for ComfyUI.
Record, clean, then give the voice a personality: monster, robot, radio, elderly… with presets you can blend, tweak and save, and a live loop that plays your changes as you make them.

--- ## ✨ Features

Voice FX node: the Dragon preset, the Grit stage open, and a live waveform

- 🎙️ **Two nodes** — **Voice Cleanup** (tidy a raw recording) and **Voice FX** (give it a personality), both plain core `AUDIO` in and out. - 🎭 **37 built-in presets in 7 categories** — Creatures, Characters, Sci-fi, Spooky, Devices, Places, Fun. - 🎛️ **Blend two presets** — mix a second preset into the current one and keep tweaking; the engine still renders once, so reverbs and pitch shifts never stack. - 🎼 **Pitch and formant are independent** — built on WORLD analysis and synthesis, so gender and age changes sound like a different person, not a chipmunk. - 👁️ **Every parameter is visible** — a friendly name up front, the technical term underneath, for the times you want an exact value. - 🔁 **Live loop** — run the workflow once, then hear every tweak on a looping region within a fraction of a second, through the same engine that renders the final output. - 💾 **User presets survive node updates** — saved presets live in your ComfyUI user directory, never inside the node package. - 📤 **Import / export** — share a preset as a `.json` file, or bring one in from a friend. ## 📦 Installation > **Requires ComfyUI ≥ 0.4.0** ### Option A — ComfyUI Manager (recommended) Open **ComfyUI Manager → Custom Nodes Manager**, search for **`Lyrebird`**, click **Install**, and restart ComfyUI. ### Option B — git clone ```bash cd ComfyUI/custom_nodes git clone https://github.com/agarzon/ComfyUI-Lyrebird.git pip install -r ComfyUI-Lyrebird/requirements.txt ``` Restart ComfyUI. > **Note:** Lyrebird depends on `pyworld-prebuilt`, which installs the `pyworld` module from prebuilt wheels. It can conflict with another node pack that requires upstream `pyworld`, since both provide the same module. ## 🎛️ Using Lyrebird Wire **Record Audio** (or **Load Audio**) → **Voice Cleanup** → **Voice FX** → **Save Audio**. Cleanup tidies the raw recording; FX gives it a personality. A ready-made version lives under **Workflow → Browse Templates → comfyui-lyrebird**: pick your own recording in Load Audio and run it. ### Voice Cleanup Runs in a fixed order: | Widget | Range / values | Default | Implementation | |---|---|---|---| | `high_pass_hz` | 0–200 (0 = off) | 80 | scipy 2nd-order Butterworth high-pass | | `hum_removal` | off / 50 Hz / 60 Hz | off | scipy iirnotch at the fundamental plus 4 harmonics | | `noise_reduction` | 0.0–1.0 | 0.5 | noisereduce spectral gating (`prop_decrease`) | | `gate_threshold_db` | -80 to -20 (-80 = off) | -80 | pedalboard NoiseGate | | `de_ess` | 0.0–1.0 | 0.0 | 4–9 kHz band split with a scipy Butterworth band-pass, compressed by a pedalboard Compressor | | `normalize_lufs` | off / -14 / -16 / -19 / -23 | -16 | pyloudnorm, then a -1 dBFS sample-peak limiter | ### Voice FX: presets and blending Pick a preset from the bar (◀ ▶ to step through, the picker to browse by category), or start from **No preset**. Every parameter stays visible and editable underneath: eight stages (Voice, Time, Tone, Grit, Modulation, Dynamics, Space, Output); every stage except Output has its own bypass toggle. - **Blend** — mix in a second preset and move the slider; your own tweaks are kept as offsets on top of the blend. - **Save (💾)** always writes a new preset into **★ Mine** — built-ins are read-only, so tweaking one only marks it "● modified", and **↺ Reset** puts it back. A name that collides with one of your own presets turns Save into an overwrite (with a warning); a name matching a built-in is rejected. - **Import / export** — Import (a `.json` file) always lands in ★ Mine; Export downloads the current settings. - **🎲** nudges a few parameters at random, if you want a starting point. ### Live loop Run the workflow once, then press **🔁 Loop** — it plays a region of the cached input on repeat, and every knob you move is heard within a fraction of a second, without re-queuing. Drag on the waveform to move the loop region (up to 30 s); **A/B** swaps between the original and the processed audio at the same playhead position. ### Where your presets live User presets are saved under `/lyrebird/presets/` — never inside the node package, so updating Lyrebird never touches them. ## 🛠️ Development ```bash uv venv -p 3.10 .venv uv pip install --python .venv/bin/python -e ".[test]" .venv/bin/pytest -q node --test tests/test-blend.mjs tests/test-controls.mjs ``` ### Smoke testing with Docker ```bash docker compose -f docker/docker-compose.yml up -d --build ``` ComfyUI will be available at **http://localhost:8188** with Lyrebird pre-loaded. Restart the container for Python changes; hard-reload the browser for JavaScript changes. ## 🐾 The story behind the name The lyrebird is an Australian bird that can imitate almost any sound it hears — other birds, camera shutters, chainsaws. A voice changer that can become almost anything felt like a fitting namesake. ## 💙 Sponsored by Wallrus ComfyUI-Lyrebird is proudly sponsored by **[Wallrus](https://wallrus.tech)**. If Lyrebird makes your ComfyUI workflow nicer, go say hi. 🐦 ## 📄 License [GPL-3.0](LICENSE) © 2026 Alexander Garzon

Name	Type	Default	Description
audio	AUDIO		—
params	STRING	{}	Effect settings as JSON. Edited by the panel.

Voice FX

Inputs (2)

Outputs (1)