LyricLike

LyricLike is a tool I built for writing song lyrics. It counts the syllables in each line, labels the rhyme scheme, and finds rhymes based on how words sound (phonemes from the CMU Pronouncing Dictionary) rather than how they are spelled.

I made this repo public mainly so I have a record of how the app actually got built.

What it does

As you type, it counts the syllables in each line and labels the rhyme scheme (A/B/C, etc). It also marks internal rhymes, which are words inside a line that rhyme with each other.

When you click a word, it finds rhymes by matching sounds, so near-rhymes show up too and not just words with the same ending. Common words come first. Names, rare words and crude words come last.

Accounts are optional. You can sign in with Google, or with an email and password, to save your pages and sync them across devices.

Stack

• Frontend: plain HTML/CSS/JS, no build step
• Backend: Cloudflare Pages Functions
• Database: Cloudflare D1 (SQLite) for accounts and saved pages
• Auth: Google OAuth, plus email and password
• Email: Resend, for sign-up confirmation and password reset links
• Bot checks: Cloudflare Turnstile on the password forms

Running it locally

You'll need Node, a Cloudflare account and Wrangler. I run Wrangler through npx at the same version the deploys use, since older global installs have crashed on me:

npx [email protected] d1 migrations apply lyricalmiracle_db --local
npx [email protected] pages dev .

Before that, create a .dev.vars file in the project root:

GOOGLE_CLIENT_ID=your-client-id
GOOGLE_CLIENT_SECRET=your-client-secret
OAUTH_REDIRECT_URL=http://localhost:8788/api/auth/google/callback

You can get these from a Google Cloud OAuth client (type "Web") with that redirect URI added.

Email and password accounts stay off unless you also set these:

PASSWORD_AUTH_ENABLED=true
PASSWORD_PEPPER=32-or-more-random-bytes-base64
APP_BASE_URL=http://localhost:8788
TURNSTILE_SECRET_KEY=your-turnstile-secret
RESEND_API_KEY=your-resend-key
EMAIL_FROM=LyricLike <noreply@your-domain>

PASSWORD_AUTH_ENABLED works as the kill switch. If you leave it unset, only Google sign-in shows up. Every link that gets emailed is built from APP_BASE_URL. The public Turnstile site key is in wrangler.toml.

You can also set ALERT_EMAIL to an address that should get a heads up when a day's signup or account email reaches 70% of its cap in functions/_email.js. Once a cap is reached, the signup, forgot-password and set-password requests answer 503 until the next day.

Preview deploys use their own database (lyricalmiracle-db-preview), so a branch never touches real accounts.

Rhyme pages

/rhymes/ holds static "words that rhyme with X" pages for search engines. They are generated from the same rhyme logic as the editor. Since the site has no build step, the pages are committed. To regenerate them:

node tools/rhymes/collect-popular-words.mjs   # rarely: re-rank popular searches
node tools/rhymes/build-pages.mjs             # rewrite /rhymes/ and sitemap.xml

The editor and the pages share one copy of the rhyme logic, rhyme-core.js. The pages leave out names, rare words and crude words, while the editor lists them last. The pages also leave out the words in tools/rhymes/page-excluded-words.json, which ad networks treat as sensitive. Those words don't get a page either, but the editor still lists them.

Word lists

These are all at the repo root, and all plain JSON arrays:

• cmudict.json: pronunciations, from the CMU Pronouncing Dictionary
• english-words.json: words that rhyme results can list, commonest first (the order is the frequency ranking)
• name-words.json: words English only spells with a capital ("wayne"), listed last. Rebuild with node tools/build-name-words.mjs
• crude-words.json: swear and sex words, listed last and never on a rhyme page
• blocklist.json: slurs. These are never listed, and typing one doesn't look anything up

Tests

The rhyme logic and the API's access checks have tests. They need Node 22 or later, and nothing else installed:

node --test "test/*.test.mjs"

GitHub Actions runs them on every pull request.

License

MIT, see LICENSE. The bundled CMU Pronouncing Dictionary has its own license, see NOTICE.