Which voice sounds most like you

October 30, 2026

Blind test. Same words, read by your real voice and four AI voices, A to D. A is the same voice in every section, and so on. Each clip is about 12 seconds. Pick the letter that sounds most like you, then open the reveal at the bottom.

1. A GTA line (you read these words in the GTA voiceover)

So Rockstar is not short on cash, but Jason and Lucia will be. Simply completing missions won't give you enough money, Nelson says, and his own recent playthrough took about eighty hours.

Real youyour own recording of these words

Voice A

Voice B

Voice C

Voice D

2. An interview answer (your words on Gent's Talk)

And in person, when people meet me, they're like, oh, you're kind of like calm. But I feel like the energetic side is like a part of me, just not all of me.

Real youyour own recording of these words

Voice A

Voice B

Voice C

Voice D

3. A shorts line (no real take exists)

I bought the cheapest phone on Amazon and the most expensive one, then dropped both of them off my roof. One of them survived. The other one is now a very expensive coaster.

Voice A

Voice B

Voice C

Voice D

Reveal which voice is which

A	Clone from the 4 interviews instant clone, from your solo speech in the 4 YouTube interviews
B	Hafu V1, the voice in every AI voiceover now professional clone, trained on 13.7 min of you reading a generic script
C	Clone from your GTA read instant clone, from the GTA 6 voiceover you recorded Oct 29
D	Clone from Hafu V1's training read instant clone, from the same audio Hafu V1 learned from (shows what the professional training adds)

What I measured

A voice-matching model compared each AI read with your real read of the same words (100 = the same recording).

GTA line: C (GTA read) 94, B (Hafu V1) 81, A (interviews) 80, D (V1's read) 80.

Interview answer: D (V1's read) 84, B (Hafu V1) 84, A (interviews) 84, C (GTA read) 80.

These scores mostly hear the mic and room, not just you. Your own real takes score only 76 to 83 against your other recordings, and every voice here lands in or near that range on a line from a different setup. The one clear standout: C, the clone from your GTA read, matches your real GTA take at 94, the others 80 to 81. A clone copies the setup it learned from, so it sounds like your GTA setup and drops on the interview answer.

Instant clones also swing with which 2.5 minutes they get: re-cutting the interview clone's sample moved its interview-answer score from 90 to 84. Treat gaps of a few points as noise. Your ears pick the winner.

Things you may hear:

A (interviews) reads the interview answer about 2 semitones higher than you did (163 Hz vs your 142 Hz).
C (GTA read) reads the interview answer about 5 semitones higher than you did (185 Hz vs your 142 Hz).
D (V1's read) reads the interview answer about 4 semitones higher than you did (184 Hz vs your 142 Hz).
Every AI read of the interview answer is 21 to 25% faster than you said it. That's the model's pace, not the clone.

How the clones were made

Each instant clone got 2.5 minutes cut the same way from its source: only your speech, long pauses shortened, same loudness. Found in total: GTA read 27.1 min (flubs and retakes left in, as you said), interviews 72.9 min (other people, music intros and crosstalk cut out), Hafu V1's training read 11.0 min of speech (the 13.7 min file minus pauses). Your real takes on this page were kept out of every clone. Same settings for every read (Eleven v4, the voiceover settings we use now). None of this changed Hafu V1.

Your call

Tell me which letter sounds most like you. Then:

Use the instant clone for GTA voiceovers: possible today with no change to Hafu V1. I'd test it on one full video's voiceover first, because instant clones can drift on long reads.
Train a professional voice on the winner's full audio. Your plan has one professional slot and Hafu V1 is in it, so this replaces it (the Scale plan has 3 slots but costs more). You'd read one sentence on screen to verify, then training takes 3 to 24 hours (ElevenLabs gives both ranges). ElevenLabs asks for 30+ minutes from one consistent setup: the GTA read has 27 (Hafu V1 was trained on 13.7), the interviews have 73 but from 4 different setups.