lyrsmith 0.1.0

TUI app for transcribing and editing synced song lyrics

lyrsmith

An AI-powered terminal UI for transcribing and editing time-synced song lyrics in LRC format.

Point it at a directory, load a track, run Whisper to get a rough transcription, then nudge timestamps until they're right. Saves directly to the audio file's tags.

Features

• AI transcription via faster-whisper — runs locally, no cloud
• Waveform display for visual timing reference
• LRC timestamp editing — stamp, nudge (fine/medium/rough), merge, split, delete, undo
• Plain text mode for unsynced lyrics
• Reads and writes lyrics tags on MP3, FLAC, OGG, OPUS

Requirements

• Python 3.13 (3.14 is not yet supported due to a ctranslate2 bug)
• libmpv (audio playback)
• FFmpeg libraries (waveform decoding via PyAV)

On Fedora/RHEL:

dnf install mpv-libs ffmpeg-free

On Debian/Ubuntu:

apt install libmpv-dev ffmpeg

On macOS (experimental):

brew install mpv

GPU-accelerated transcription is not available on macOS; transcription runs on CPU.

NVIDIA GPU — if you want GPU-accelerated transcription, also install libcublas and make sure it is registered with ldconfig:

# Fedora example — package name may vary
dnf install libcublas
echo "/usr/local/cuda-12.0/targets/x86_64-linux/lib" | sudo tee /etc/ld.so.conf.d/cuda.conf
sudo ldconfig

Install

From source using uv:

git clone https://github.com/triluch/lyrsmith
cd lyrsmith
uv tool install . --python 3.13

Or with pipx:

pipx install .

Updating

cd lyrsmith
git pull
uv tool install . --python 3.13 --reinstall

Usage

lyrsmith [DIRECTORY]

If no directory is given, the last-used directory is restored (or the current working directory on first run).

Press F1 inside the app to see all keybindings.

Recommended model: large-v3-turbo (the default) gives good results at reasonable speed. large-v3 is noticeably more accurate for tricky lyrics but slower and needs more RAM. Configure via F2.

Voice Activity Detection (VAD) is disabled by default. If you're getting a lot of repeated lines or hallucinated text in the transcription, try enabling it with a low threshold like 0.00005–0.0001 — this pre-filters the audio before Whisper and can significantly improve results on tracks with instrumental sections or background noise.

---

And yeah, of course it is slopped out, what did you expect in current times?