Lyrebird v0.3.1

Download Lyrebird-win64.zip below (about 36 MB), unzip it, and run Lyrebird\Lyrebird.exe. Guide with screenshots: HOWTO.md.

This release has everything from v0.3.0: the redesigned light and dark interface, sessions, text loading and drag and drop. On top of that, it fixes issues found in a code review:

• Theme: the switch starts on System and keeps following Windows. Before, closing the app could fix it on Light or Dark.
• Sessions:
  • They restore "(Built-in voice)" and the Windows default microphone.
  • A session whose voice is missing falls back to the built-in voice and names the missing voice.
  • A damaged session file is rejected before anything changes, so it can't overwrite the session you have open.
• Controls:
  • Greyed-out sliders no longer move with the mouse wheel.
  • Ctrl+O and Ctrl+S work with Caps Lock on, and Shift picks Save as.
  • ▶ follows the voice a session selects.
• Trim: Auto-trim keeps the 15 s limit for long clips.
• Robustness:
  • Dropping an audio file while recording or rendering waits instead of interrupting.
  • Open and Save still work when the output drive is unplugged.
  • Bold speaker names resize correctly on monitors with different scaling.
  • Setup recovers from a half-created environment folder.

The first launch after updating runs a short setup to add the new interface packages.

The exe isn't code-signed yet, so Windows SmartScreen may warn you. Click More info → Run anyway.

Lyrebird v0.3.0

Use v0.3.1, which adds a few fixes on top of this release.

New in 0.3

• A redesigned interface. It's calm and roomy, uses the Lyrebird brand colours, and has light and dark modes. The Light / Dark / System switch follows your Windows theme or is set by hand. The script gets the big area; Voice, Delivery and Output sit in cards with small colour markers; Render is the one solid button.
• Sessions: Save (Ctrl+S) stores the script and every setting in a .lyrebird file, and Open... (Ctrl+O) brings it back. It also opens .txt and .md scripts.
• Drag and drop: drop a script, a session or an audio clip anywhere on the window. An audio clip starts adding it as a voice.
• The first-run setup window matches the new look.
• The HOWTO has been redone with new screenshots, including dark mode.

Fixed

• Updating while Lyrebird is open no longer fails with "Access is denied", and setup errors now say what actually went wrong.
• Setup no longer re-runs when switching between the release and a source checkout.

Lyrebird v0.2.0

Download Lyrebird-win64.zip below (about 35 MB), unzip it, and run Lyrebird\Lyrebird.exe. Guide with screenshots: HOWTO.md.

If you already have v0.1.0, the first launch runs a short setup again to add one new package. After that it starts straight away.

New

• Splash screen with the new Lyrebird brand, the version and loading status. The exe, window and taskbar now use the Lyrebird feather icon.
• Recording sources: besides microphones, a What you hear source records whatever your PC is playing on any output, so you can capture a voice from a video or call.
• Trim before saving: each recording or import opens a waveform view with draggable start and end handles, Auto-trim and Play selection. You can cut a clean 10 seconds out of a long clip.
• Manage voices: rename or delete voices, and open the voices folder, from inside the app.
• Per-voice settings: give a voice its own exaggeration, CFG/pace or temperature. It's used whenever that voice speaks, including in dialogues.
• Output formats: WAV 16-bit, WAV 32-bit float or FLAC 24-bit, at 24, 44.1 or 48 kHz.
• Stop cancels a long render after the current chunk.
• Remembers your setup: model, sliders, seed, format, voice, source, output folder and window size.
• An example dialogue script in examples/.

Fixed

• Play is greyed out until there's a voice sample to play, and Play output until you've rendered something.
• A microphone that won't open through WASAPI is retried through MME and DirectSound.
• A silent recording is rejected with an explanation instead of being saved as a voice.

Notes

• The exe isn't code-signed yet, so Windows SmartScreen may warn you. Click More info → Run anyway. Signing through SignPath Foundation is being set up.
• Every generated file carries Resemble AI's imperceptible Perth watermark. Only clone voices you have permission to use.

Lyrebird v0.1.0

First release of Lyrebird, a small Windows app for voice cloning with Chatterbox TTS.

Download Lyrebird-win64.zip below (about 35 MB), unzip it, and run Lyrebird\Lyrebird.exe. Step-by-step guide with screenshots: HOWTO.md.

What it does

• Record 10 s from any microphone, or import wav/mp3/flac/ogg. Voices are saved by name in a voice library.
• Type any amount of text and click Render to get a 24 kHz mono WAV.
• Multi-speaker dialogue: start a line with a saved voice's name (Alice: Hi!) to switch speaker.
• Three models: Standard (English), Turbo (English, fast, with [laugh]/[sigh]-style sound tags) and Multilingual (23 languages).
• Sound tags are highlighted as you type, and an Insert tag menu lists all 19.
• Options: exaggeration, CFG/pace, temperature and seed.

First run

The zip doesn't include PyTorch, which is why it's 35 MB and not 5 GB. On first launch, a setup window downloads Python, PyTorch and Chatterbox into %LOCALAPPDATA%\Lyrebird: about 3 GB with an NVIDIA GPU, needing about 6 GB of disk space. The window shows progress. Setup picks the CUDA build that matches your driver, or the CPU build if there's no NVIDIA GPU. Each model is then downloaded (3 to 4 GB) the first time you render with it.

Requirements

64-bit Windows 10 or 11. An NVIDIA GPU is strongly recommended (about 6 GB of VRAM); CPU-only works but is slow.

Notes

• The exe isn't code-signed yet, so Windows SmartScreen may warn you. Click More info → Run anyway.
• Every generated file carries Resemble AI's imperceptible Perth watermark. Only clone voices you have permission to use.