Getting a song’s lyrics using AI

@nixtoshi
3 min readNov 1, 2024

--

I was listening to a song on Spotify, but they didn’t have the lyrics for it like it’s usual for many newer or niche songs. So I generated them using AI, the details of how I did it are at the bottom of the page.

The song and the lyrics:

Lyrics of KOKA KOLA by Ely Oaks

She’s like Madonna, she’s loca,
She’s sippin’ Koka Kola, in Gabbana
She’s loca, she’s sippin’ Koka Kola
Like Madonna, I want her, more than any dollar
In Gabbana, she’s loca, she’s sippin’ Koka Kola

She’s like Madonna, she’s loca,
She’s sippin’ Koka Kola, in Gabbana
She’s loca, she’s sippin’ Koka Kola
Like Madonna, I want her, more than any dollar
In Gabbana, she’s loca, she’s sippin’ Koka Kola

Ko-ka Ko-la
Ko-ka Ko-la
Ko-ka,
Sippin’ on Ko-la, wanting that — Ko-ka
Sippin’ on Ko-la

She’s like Madonna,
She’s loca, she’s sippin’ Koka Kola, in Gabbana
She’s loca, she’s sippin’ Koka Kola
Like Madonna, I want her, more than any dollar

In Gabbana, she’s loca,
She’s sippin’ Koka Kola

[background voice]
Sippin’ Koka Kola, like Madonna,
She’s sippin’ Koka Kola, like Madonna
More than any dollar, like Madonna
She’s sippin’ Koka Kola
[background voice ends]

She’s like Madonna, she’s loca
She’s sippin’ Koka Kola, in Gabbana
She’s loca, she’s sippin’ Koka Kola
Like Madonna, I want her more than any dollar
In Gabbana, she’s loca, she’s sippin’ Koka Kola

Kola, Kola, Kola, Kola, Kola
Kola, Kola, Kola, Kola, Kola

Kola, Kola, Kola, Kola, Kola
Kola, Kola, Kola, Kola, Kola

Kola, Kola, Kola, Kola, Kola
Kola, Kola, Kola, Kola, Kola

Sippin’ Koka Kola

She’s like Madonna, she’s loca
She’s sippin’ Koka Kola, in Gabbana
She’s loca, she’s sippin’ Koka Kola
Like Madonna, I want her, more than any dollar
In Gabbana, she’s loca, she’s sippin’ Koka Kola

Ko-ka Ko-la
Ko-ka Ko-la
Ko-ka,
Sippin’ on Ko-la, wanting that — Ko-ka
Sippin’ on Ko-la

How I got the lyrics:

  • I downloaded the song as MP3 using the yt-dlp command-line utility on Mac.
  • I used OpenAI’s Whisper V3 on Replicate.com to transcribe the song from .MP3 to .txt (it did an OK job). The price was around $0.0128 for each minute that I rented the machine doing the job, it run for 2 minutes for a total cost of around $0.025.
  • I gave ChatGPT-4o (latest) the transcribed lyrics to improve and correct them using multiple prompts. It did an OK job and the lyrics were easier to read since Whisper doesn’t sepparate what it hears into paragraphs, it just dumps all the words into the output without good spacing, or punctuation marks. Some portions of the song were almost perfect after this step, but ChatGPT changed the length of the lyrics significantly.
  • Then I manually edited ChatGPT-4o’s and Whisper’s output as I listened to the song, and made some aesthetic changes like replacing all “Coca-cola”s to “Koka Kola”, the name of the song.
  • All in all, I was fairly impressed at how quicker I was able to transcribe these lyrics after failing to find them online, or on Spotify.

--

--

@nixtoshi
@nixtoshi

Written by @nixtoshi

My site: nixtoshi.com @nixtoshi on Twitter. I coordinate the Spanish translation of bitcoin.org. Interested in crypto, anti-aging and type 1 civilizations