The Day I Read My Own Lyrics Wrong on Genius
Someone transcribed a DARK I track on a lyrics site and got a whole bar inverted. Not a typo — a meaning flip. The line was about outwitting the thing chasing you; their version had it the other way around. Thousands of people could read that as canon, and I had no edit button that actually meant anything. That's the moment it clicked: I don't own the text of my own catalog. I own the masters, the stems, the distribution — but the words, the annotations, the why-behind-the-bar — that lived on a platform I didn't control, scraped and approximated by strangers.
So I built my own. Not a vanity wall of lyrics with a nice font. An annotated, source-verified lyrics archive — a Genius I actually own — and I'm going to argue it's one of the most underrated pieces of catalog infrastructure an independent artist can build.
Lyrics Are Data, Not Decoration
Here's the reframe that matters. Most artists treat lyrics as marketing collateral — you slap them in an Instagram caption, maybe a lyric video, then forget them. But a lyric line is a structured record. It has a timestamp in the audio. It has a section (verse, hook, bridge). It has a confidence level — was that definitely the word, or did you mumble it? It has provenance — which session, which take, which version. And it has annotation — the reference, the double-meaning, the sample, the bar you're proudest of.
When you store all of that, lyrics stop being decoration and become a queryable layer of your catalog. DARK I — Outwitting the Devil — maps each of its ten chapters to Napoleon Hill's 1938 manuscript. That mapping IS metadata. "Kether" sitting at the top of the Kabbalistic Tree of Life is a concept that threads through specific lines across the record. If that lives only in my head, it dies with my memory of it. If it lives in a structured archive, it becomes searchable, linkable, and — critically — survivable. A catalog that can explain itself outlives the artist's recall.
Why This Is Infrastructure, Not Ego
Call it vanity if you want, but vanity doesn't pay licensing. Here's what an owned lyric archive actually does for the business:
- Sync and licensing. When a music supervisor wants a line for a trailer, the first thing they need is clean, confirmed lyrics with timecodes. If you can hand them a source-verified document in thirty seconds, you look like a label. If you're squinting at a streaming caption, you look like a hobby.
- Publishing accuracy. Your registered lyrics with your PRO and the words people actually hear should match. Discrepancies cost you in disputes. An archive is your evidence.
- SEO and discovery you control. "DAJAI [line] lyrics meaning" should land on your page, not a scraper's. Owning the annotation means owning the search result.
- AI-readiness. Every model that ingests your catalog reads the text it can find. If the only text out there is a bad transcription, that's the version that gets embedded into the machines. Feed the machines your version.
This is the same logic as owning your masters or shipping stems. You're not building a shrine. You're building a moat.
The Pipeline: Whisper First, Hands Second
The honest part: doing this by hand for 14,000+ catalog tracks would take a lifetime. So the archive runs on a two-pass system, and the order is the whole trick.
Pass one — Whisper does the grunt work. I run each track through Whisper for a first-pass transcript with word-level timestamps. This is not the final text. Whisper is a tireless intern, not a poet. It will hear "wataa" as "water," it'll fumble adlibs, it'll punctuate like it's writing a press release, and it has zero idea which words are load-bearing. But it gets me 80% of the way to a timed skeleton in minutes per track, and timestamps are the expensive part to produce manually.
Pass two — hand-correction by the person who wrote it. This is non-negotiable and it's where "source-verified" earns the name. I go line by line against the actual master — not the Whisper guess, the audio — and I fix the words, restore the intentional misspellings, mark the adlibs, and flag anything I'm genuinely unsure I enunciated. The output of pass two carries a verified: true flag and a date. That flag is the entire value proposition. Anyone can scrape a guess; only the artist can confirm the truth.
The same discipline I use mastering applies here: the machine drafts, the human with the ears decides. Whisper's confidence score is a flag, not a mandate — a low-confidence word tells me where to listen harder, never what to write.
What a Single Entry Actually Looks Like
Each line in the archive is a small record, not a string in a text file. In practice it carries: the text (corrected), the start/end timestamp from Whisper, the section it belongs to, a machine-confidence number from the first pass, a verified boolean from the second, and an optional annotation — the reference, the production note, the Hill chapter it ties to.
That structure is what makes it infrastructure instead of a Notes file. I can query "every line tagged to Tree-of-Life imagery across DARK I." I can export clean lyrics-only for a publisher, or lyrics-plus-annotation for a fan-facing page, or lyrics-plus-timecode for a sync request — three different products from one source of truth. And because the production notes (which beat, mastered by Solana Conejo, which stem revision) live in the same record as the words, the archive doubles as liner notes that never get lost.
Own the Source of Truth Before Someone Else Defines It
The DARK Library is a planned ten-volume cycle. That's potentially a decade of work where the meaning is the point — chapters mapped to a 1938 manuscript don't explain themselves to a casual listener. If I let third-party sites become the canonical record of what those lines say and mean, I've handed away the interpretation of my life's work to whoever typed fastest.
Building your own annotated archive isn't about ego. It's about being the primary source for your own catalog. Masters, stems, lyrics, annotations — same principle, four layers deep. Whisper gets you the skeleton cheap. Your ears make it true. And once it's true and yours, it pays you in licensing, in search, in accuracy, and in legacy for as long as the catalog exists.
FAQ
Do I need to be technical to build a lyrics archive like this?
Not as much as you'd think. Whisper has free, GUI-friendly versions, and the "archive" can start as a structured spreadsheet — one row per line with columns for timestamp, text, confidence, verified, and notes. The discipline matters more than the tooling: machine first pass, human second pass, and a date on every verification.
Why not just use Genius or a lyrics site?
Because you don't own it, can't guarantee accuracy, and don't control how it's used or scraped. Third-party sites are a distribution surface, not a source of truth. Build the canonical version yourself, keep it verified, and syndicate out to those platforms — never let them be your only record.
Is Whisper accurate enough to skip hand-correction?
No, and skipping the second pass defeats the entire purpose. Whisper is excellent at timing and decent at common words, but it mangles slang, intentional misspellings, adlibs, and anything genre-specific. The "source-verified" claim only holds because the artist confirms every line against the master audio. The machine drafts; the human decides.
How does this actually make money for an independent artist?
Three direct paths: faster, cleaner sync/licensing pitches because supervisors get timecoded confirmed lyrics instantly; fewer publishing and PRO disputes because your registered and heard lyrics match with evidence; and search traffic you own instead of donate to scrapers. It's the same return profile as owning masters — slow, compounding, and entirely yours.