My home-renovation videos are edited by a person who knows my recordings well, including the words I sometimes do not pronounce clearly. When he heard this AI narration, he said it sounded better than my own recording. That was his reaction to this result, not a general comparison of every AI voice.
Test a short passage before making the full English version
After finishing the Chinese video for 970 LAB, I wanted to try an English version while keeping my own voice rather than switching to a different narrator.
The first English sample already sounded somewhat like me, but the delivery was stiff and followed a Chinese rhythm. I did not generate the whole script immediately. I kept testing short passages. After changing the generation approach, the delivery sounded more natural to me, so I used that direction for the complete narration.
A familiar voice speaking English more fluently than I can
Listening to the finished English version gave me a very direct reaction: I could not record it that fluently myself. That is a comparison with my own speaking, not a claim about an English level or a language assessment.
The narration, captions and on-screen text were prepared separately in Chinese and English. Both versions tell the same story, while each one remains complete in the language its audience sees and hears.
The result worth keeping
The most useful result is that the two finished clips can be heard side by side. In this experiment, my authorised voice profile could produce complete videos in two languages. Whether it sounds like me and whether the delivery feels natural remain judgements I made after listening, not a similarity score.
The article and videos use an AI-generated version of my voice with my permission. The original recordings, voice model and speaker data remain private.