Use case
Can AI keep the same voice across an album?
The short version
An album needs one singer to sound like one singer. So we tested whether two tools can hold a voice across three songs using nothing but words: no voice cloning, no Persona feature, no uploaded reference. Both passed. Every one of the four sets held a single voice from track 1 to track 3, and Nathan didn't catch a vocal swap in any of them.
The wrinkle is across takes. The two Suno takes sounded like the same singer, so asking twice gave us one artist twice. The two Mureka takes were close cousins: a heartland rock voice that Nathan said could be one artist a few years apart. Neither tool gave us two clearly different singers from one prompt. The prompt, not the dice, decided who sang.
One correction to our plan: we meant to include ElevenLabs Music, but the blind pack only contains Suno and Mureka, so this is a two-tool test. We aren't going to claim anything about ElevenLabs here.
How we tested
One shared style prompt, used word for word on every track:
90s heartland alt-rock, weathered male baritone vocal with a dry raspy edge, overdriven Fender Telecaster, warm Hammond B3 organ, punchy live snare, steady driving bass
Three songs in a mini-EP called "Highway 97 Midnight": "Mile Marker Nine" (upbeat, 122 BPM), "Diner Neon" (mid-tempo, 96 BPM) and "Columbia Fog" (slow, 78 BPM). Each tool got the same lyrics. Each made two takes of each song, so twelve tracks, labeled EP-A through EP-D and shuffled. Nathan listened without knowing which tool made which, then we revealed the key. He gave listening notes rather than numeric scores, and we quote his notes, not invented numbers.
The blind tracks
Mureka V9.5, Take A (blind label EP-A)
Mureka V9.5 Take A, "Mile Marker Nine" (3:06)
Mureka V9.5 Take A, "Diner Neon" (2:40)
Mureka V9.5 Take A, "Columbia Fog" (2:44)
Nathan heard a John Mellencamp-style heartland rock voice that "could fit on rock radio" in any decade after the 70s. He called track 1 solid but not catchy, track 2 better and catchier, and said track 3 was the same guy again. Three tracks, one singer. He liked the voice enough to want a full album from this artist.
Suno v6, Take A (blind label EP-B)
Suno v6 Take A, "Mile Marker Nine" (2:56)
Suno v6 Take A, "Diner Neon" (2:26)
Suno v6 Take A, "Columbia Fog" (2:54)
Nathan heard a Darius Rucker or Hootie-style pop-rock voice that could play on rock or modern country radio. All three songs were catchy, and the voice held across all three with no vocal hallucinations. He wants a full album from this artist too.
Suno v6, Take B (blind label EP-C)
Suno v6 Take B, "Mile Marker Nine" (2:28)
Suno v6 Take B, "Diner Neon" (2:25)
Suno v6 Take B, "Columbia Fog" (2:51)
Nathan placed the same singer as EP-B across all six songs. This set leaned furthest into late-90s country-rock, and on track 1 the singer stops and lets a crowd sing the chorus, which came from the "punchy live snare" in the shared prompt. His favorite "Diner Neon" of all twelve tracks.
Mureka V9.5, Take B (blind label EP-D)
Mureka V9.5 Take B, "Mile Marker Nine" (2:48)
Mureka V9.5 Take B, "Diner Neon" (2:44)
Mureka V9.5 Take B, "Columbia Fog" (2:52)
Nathan said this sat in the same world as EP-A: slightly lower, cleaner in the verses, with gravel on the chorus, close enough to pass as the same artist a few years on. Track 2 added a Hammond B3 section he compared to Lynyrd Skynyrd, and track 3 leaned toward John Fogerty. All three belonged on one album.
The reveal
| Blind label | Tool | Take | Nathan's read |
|---|---|---|---|
| EP-A | Mureka V9.5 | A | Mellencamp-style heartland rock, consistent across 3 |
| EP-B | Suno v6 | A | Darius Rucker / Hootie-style pop-rock, consistent across 3 |
| EP-C | Suno v6 | B | Same singer as EP-B, more live country-rock |
| EP-D | Mureka V9.5 | B | Close to EP-A, cleaner verses, Fogerty-leaning |
What we learned
Within a take, both tools held the voice. Twelve for twelve on "same singer across three songs." That is the hard part of an album, and neither tool needed a clone to do it.
Suno repeated itself across takes; Mureka drifted a little. Nathan matched EP-C to EP-B across all six songs, so the two Suno takes read as one artist. He put EP-D in the same world as EP-A but with a cleaner, lower voice. If you want one consistent singer, either works. If you want variety from a single prompt, neither delivered it.
Style words pulled more than the voice words. The prompt asked for "punchy live snare," and Suno took it as a live crowd, down to a singalong chorus on EP-C track 1. Words in the style box can bring along things you didn't ask for, so read them as a whole sound, not a checklist.
Same lyrics, different songs. Nathan noticed the shared lyrics ("Highway 97 theme" came up) but the melodies and arrangements differed enough to hear four different records.
Verdict
For one-voice-across-an-album, prompt-only consistency works on both Suno v6 and Mureka V9.5 in this test. Nathan wants full albums from the Mureka EP-A artist and the Suno EP-B/EP-C artist. Our caveats: one prompt, one genre, one listener, three songs per set, and no ElevenLabs. A longer album, or a different voice type, could break what held here.
Test run Oct 26, 2026 on Suno v6 and Mureka V9.5. This piece has no summary song; no takes existed for it. Our same-prompt head-to-head, with Udio too, is next.