Sing Along

by Devon Lin
5
4
3
2
1
Score: 50/100

Description

Obsidian plugin: sync Japanese lyrics with a YouTube video, auto-align locally, click to jump and loop lines.

Reviews

No reviews yet.

Stats

0
stars
2
downloads
0
forks
1
days
1
days
1
days
0
total PRs
0
open PRs
0
closed PRs
0
merged PRs
0
total issues
0
open issues
0
closed issues
2
commits

Latest Version

2 days ago

Changelog

First release. Install with BRAT: neonthewei/obsidian-sing-along. Requires macOS Apple Silicon, uv and ffmpeg.

README file from

Github

Sing Along

Practise Japanese songs in Obsidian. Sing Along pairs a note of three-line lyrics (romaji, Japanese, translation) with a YouTube video, aligns every line to the video's audio on your own machine, and lets you click a line to jump, follow along, and loop a line or a section.

中文說明

Desktop only, macOS Apple Silicon only (for now). Audio alignment runs locally with MLX Whisper. No API key and no cloud transcription service.

Install

Not in the community store yet. Install with BRAT:

  1. Install and enable BRAT from Community plugins.
  2. Command palette → BRAT: Add a beta plugin for testing → paste neonthewei/obsidian-sing-along.
  3. Enable Sing Along in Settings → Community plugins.
  4. In a terminal: brew install uv ffmpeg.

BRAT keeps the plugin updated when a new release is published.

Let an AI assistant walk you through it

This repo includes an agent skill, skills/sing-along, that guides you step by step: checks your Mac, installs the plugin, creates an empty lyric note, runs the first alignment and fixes common errors. It never fetches or writes song lyrics; you paste those yourself.

npx skills add neonthewei/obsidian-sing-along

Or copy it by hand for Claude Code:

git clone https://github.com/neonthewei/obsidian-sing-along /tmp/sing-along
mkdir -p ~/.claude/skills && cp -r /tmp/sing-along/skills/sing-along ~/.claude/skills/

Then tell your assistant: "Help me set up Sing Along".

Features

  • Opening a lyric note (one with a sing_along_youtube property) opens the player view automatically. Ordinary notes are untouched.
  • One tab per song. Reopening a song returns to its tab with playback position and loop state kept.
  • Auto-alignment: the first time you use a video, its audio is analysed and a hidden timestamp is written above each line. No manual tapping.
  • Click a line to jump, current line is highlighted and kept centred, section dock on the left, playback dock at the bottom (speed, previous, play/pause, next, loop line).
  • The video scrolls with the page and shrinks into a mini player above the section dock when it leaves the screen.

Keyboard

Key Action
Space Play / pause
↑ / ↓ Previous / next line
R Toggle line loop
⌘R Toggle section loop
⌘↑ / ⌘↓ Previous / next section

Keys are ignored while you type in a text field or use the speed menu.

Lyric note format

---
title: Song title
sing_along_youtube: https://www.youtube.com/watch?v=VIDEO_ID
---

## 歌詞對照

### Verse 1

*Romaji line*  
**日本語の行**  
Translation line

## Other notes

Only the block under the ## 歌詞對照 heading is read, up to the next level-2 heading. ### headings become sections. Repeated lines in different sections get independent timestamps. After alignment each line gets a hidden comment:

%% sing-along: 00:12.340 --> 00:17.820 %%

You supply the lyrics yourself. The plugin does not fetch or ship any lyrics.

Requirements and what the plugin does on your machine

Please read this before installing.

  • External programs. Alignment runs a bundled Python script (align.py) through uv as a child process. You need uv and ffmpeg installed (for example brew install uv ffmpeg).
  • Downloads on first use. uv installs the Python packages pinned in align.py (stable-ts[mlx], yt-dlp), and the speech model mlx-community/whisper-large-v3-turbo-q4 (about 450 MB) is downloaded from Hugging Face once.
  • YouTube audio. yt-dlp downloads the audio of the public video you choose so it can be analysed. It does not use browser cookies or your account. Private, restricted or non-embeddable videos fail with a clear error.
  • Local cache. The Python worker is bundled inside main.js and written to ~/.cache/obsidian-sing-along/worker/ on first use. Audio, the model and alignment results are stored in ~/.cache/obsidian-sing-along/, outside your vault. Delete that folder to reclaim the space.
  • What goes into your vault. Only the video URL (frontmatter) and the timestamp comments in your lyric note.
  • Playback. The player uses the official YouTube IFrame API inside an isolated webview. YouTube's own controls and ads remain.

Known limitations

  • Timestamps are model estimates. Long instrumentals, harmonies, repeated lines and unusual vocal styles can drift. If alignment fails or the line count does not match, the note is not modified.
  • Looping uses YouTube seek plus ~200 ms polling: fine for practice, not sample-accurate.
  • Timestamps belong to one video. A different MV, live or album version needs its own alignment.
  • Japanese only for now (alignment language is fixed to ja). The UI is in Traditional Chinese.

Development

npm install
npm test                 # JavaScript + Python unit tests
npm run build            # bundles into dist/
OBSIDIAN_PLUGIN_DIR=/path/to/vault/.obsidian/plugins/sing-along npm run install-plugin

Source lives in src/. align.py is a JSON-over-stdio worker: input videoId and lines, output progress events and per-line start/end times.

References

License

MIT


中文說明

Sing Along 讓你在 Obsidian 裡跟著 YouTube 練唱日文歌:筆記裡放「羅馬拼音/日文/中文」三行歌詞,插件在本機分析影片音訊,自動對上每一句的時間,之後可以點句跳播、跟隨標亮、單句或整段循環。

  • 安裝:先在社群插件裝 BRAT,命令面板執行「BRAT: Add a beta plugin for testing」,貼上 neonthewei/obsidian-sing-along,再到設定啟用 Sing Along。
  • 想讓 AI 一步步帶你裝:執行 npx skills add neonthewei/obsidian-sing-along,再跟 Claude Code 說「幫我設定 Sing Along」。這個 skill 不會幫你抓或寫歌詞,歌詞要自己貼。
  • 只支援桌面版,目前只支援 macOS Apple Silicon。
  • 需要先安裝 uv 與 ffmpeg;第一次使用會下載 Python 套件和約 450 MB 的語音模型,之後使用快取。
  • 音訊、模型與分析結果放在 ~/.cache/obsidian-sing-along/,不進 vault;vault 裡只會多影片網址和每句上方的隱藏時間註解。
  • 歌詞要自己準備,插件不會抓取或附帶任何歌詞。
  • 時間點是模型估計,特殊唱腔、合聲或長間奏可能有偏差;對齊失敗時不會改動筆記。