š§Ŗ The testĀ
I wanted to give everyone the same starting point to make the comparison fair.
š Voice trainingĀ
I then needed a script for my voice clones to read out loud.
Suddenly, the sky pixelated. Colors drained. A low hum rose from the earth. Fluffy didnāt panicāhe remembered this feeling. The pull, the erasure, the end. Somewhere far away, a child clicked āGenerate New Story,ā and everything Fluffy loved began to dissolve.
ļ¼The audio and script are obtained from online sources.ļ¼
Thatāll be the passage I ask our voice cloning sites to reproduce.
For comparison, hereās how I sound saying it:
Now, letās hear which AI voice cloning result sounds the most like my real voice.š
š£ļø The results
My main focus is on how closely the cloned AI voice resembles mine.
But Iām also interested in things like how natural the reading feels (pacing, tone, etc.), audio quality, and so on.
Get ready to listen to a bunch of fake AI voice.
1. MiniMax
You might remember MiniMax from its impressive Hailuo AI video model that landed near the top inĀ my image-to-video test.
But it also has a speech model called speech-02-hd and a standalone audio platform for text-to-speech generation. How does it handle cloning my voice?
Oh wow, thatās surprisingly good!
Clean, crisp audio, natural pacing, and a voice that isnātĀ quiteĀ me but by far the closest weāve heard so far.
Where to try:Ā www.minimax.io/audio
2. A2E
In addition to AI voice cloning, A2E offers image generation using models.
I enjoyed its clean interface for managing cloned voices with a āGround Truth Audioā reference. But how does the clone voice?
Not bad at all. That does sound like me! Also, the pacing and narration come off as rather natural.
Where to try:Ā video.a2e.ai/voice_clone
3. Descript
Descript is a popular video creation and editing platform that now also offers voice cloning. Is it any good?
Clean, loud audio. Rather natural reading, although I wish the narrator took a breath between sentences every once in a while.
Unfortunately, the voice is only somewhat Daniel-adjacent. Iām pretty sure it wouldnāt fool anyone who knows me.
And just as with Colossyan, you have to create a video project, even if you only want the text-to-speech aspect. The free plan is quite generous, though, and lets you generate hours of audio and have multiple clones.
Where to try:Ā www.descript.com
4. Mango AI
Mango AI has a just-for-fun focus with features like talking animals, singing videos, face swap, cartoons, etc.
Howās its voice cloning?
Mango AI nails the pacing, but thatās where the positives stop.
The quality isnāt great, like āAI Danielā is talking on a badly damaged telephone line through a breathing mask. The clone doesnāt sound like me and gives off a robotic vibe with lapses in pronunciation.
Where to try:Ā mangoanimate.com
5. PlayHT
Oddly enough, despite advertising voice cloning and having a āYour Voicesā section in the dashboard, PlayHT doesnāt provide the option to clone a new voice for free accounts.
But thereās a āSpeech Editorā that lets you upload an audio sample and then edit the transcript to change what the voice says. So I used that as a workaround:
While the voice is vaguely Original-ish, the narrator sounds like he was told to finish his lines before a bomb goes off.
Thereās a good chance that using proper voice cloning instead of the āSpeech Editorā workaround would lead to better results. But as it stands, this isnāt good at all.
Where to try:Ā play.ht
