Improved bulk lyric editor shipping later this month. Users are responsible for the legal and ethical use of our outputs.
3 min read

Special Death (Mirah) - English Lyric Adaptation with Vocal Pacing

Mirah's Special Death adapted with custom English lyrics that fit the original melody, plus a frank look at why ChatGPT-written revision notes do not help on audio deliveries.

Share:

Posted by

Special Death by Mirah custom English lyric adaptation with vocal pacing adjustments

Eva came to ChangeLyric with custom English lyrics for Mirah's "Special Death." The new lines were not a one for one swap. Some had longer meter than the original and needed pacing adjustments to fit Mirah's vocal style.

Our first version missed the mark. Pronunciation was inconsistent, the phrasing did not sit on the melody, and the music itself felt altered. Eva sent a detailed revision note and we delivered a second version that addressed it.

Listen & Compare

Hear the original song and the custom version side by side

Original
0:00--:--
Custom Version
0:00--:--

Transcripts are auto-generated and may not perfectly reflect the audio.

Original Lyrics
A terrible mistake was made The seven women were left alone With faithful arrows you Might get back in the saddle Once you made it hotter The thankless, holy praise Is left alone, why bother To cast a stone in water? Of ten strong horses tried to save The one that made it through If you knew that I could take the pain
Custom Lyrics
An horrible mistake was made The seven women were left alone With faithful staves by May fly back through to the muddle Though you made it harder I'd break the spell the same It's time and no other To cast the stone in place Of ten thousand strong horses trying to save The one that made it through If you knew that I could hold the pain

The first version looked as if it had just been run through AI without checking whether the lyrics actually fit the melody, mixing different pronunciations and changing the music of the song. The second version, however, clearly shows real work after giving more details about how it should sound and with the editor paying much closer attention to detail.

Eva N.

Eva's revision note (almost certainly AI-written)

Heads up: the note below was almost certainly generated by ChatGPT or a similar LLM. The labeled-spec format, the parenthetical jargon definitions ("Flap T (where T sounds like D)"), and the framing of our work as "AI settings" with knobs to turn are dead giveaways. We are pointing this out because it matters more than it sounds, as we explain underneath the quote.

Eva's revision note, pasted verbatim:

For the lyric modifications, I'm looking for a specific "Mid-Atlantic" aesthetic (a hybrid between refined American and British) to match Mirah's exact diction. Please adjust the AI settings with the following priorities:

Articulation & "True T": All "T" sounds must be crisp, dental, and dry (referencing Mirah's "T" in "Mistake"). Avoid the standard American "Flap T" (where T sounds like D). This is crucial for "It's time"; the "T" should be sharp and intentional, not lost in the breath.

Non-Rhotic Accent: Soften the "R" at the end of words like Daughter, Harder, or Other. They should sound like open, elegant vowels rather than hard American "R"s, giving it that European/Indie-folk vibe.

Dramatic Pause & Phrasing: In the final line, please insert a brief but intense pause right after "It's time," leaving a short beat of silence before delivering "and no other." This silence should add weight and solemnity to the phrase.

Percussive Consonants: Even though the overall tone is soft and ethereal, the consonants should be precise and percussive. The vocals shouldn't sound "mushy"; I need them to be fragile yet perfectly intelligible.

Why AI-written feedback on audio deliveries does not work

We are flagging this kindly, not as a dig at Eva. The issue is the tool, not the customer.

Large language models are text only systems. They cannot actually hear the audio file you upload to them. When a customer sends our v1 delivery to ChatGPT and asks for feedback, the model is not listening to anything.

Instead, the LLM produces plausible sounding instructions based on what it knows about the artist's general style. That is why Eva's note reads like a vocal coaching script for Mirah herself rather than a reaction to anything specific in the v1 we delivered.

Some of the items happened to apply to what was actually in the file. Others did not. The model had no way to tell the difference, because it could not hear the file in the first place.

If you ever need to send revision feedback on a music delivery (to us or to anyone else), please write one or two sentences yourself describing what felt off. "The Ts sound mushy," "the second verse feels rushed," or "I cannot hear the consonants on the last line" beats a polished AI essay every time, because it is grounded in what you actually heard.

Browse more projects in our lyric swap showcase, or start your own custom lyric swap.