Thoughts on An Xiao's trans-lingual English?

I see that An Xiao is the latest voicebank to be introduced. :+1:

Any thoughts on how well he sings in English?

Kevin is currently the only native male voicebank. He’s clear, but more than a little nasal and a bit bright.

Ryo has an attractive voice, but the trans-lingual conversion isn’t that convincing, The words “a tree” sound more like “a shree”. Putting a gap between the two words makes him sing “a pea”, which is even worse. There are other issues, including:

  • The r-colored vowels tend to be non-rhotic;
  • The initial /t/ tends toward /dx/;
  • His /ih/ is more rounded, as /eh/;

So at this point, he’s really not usable for me as a lead English voice.

I think Mo Chen handles most of the vowels better than Ryo, but clearly the voicebank is trained on the same sample set, as he has the same issues with "a tree as Ryo does.

I suspect that An Xiao will have the same sort of issues as other trans-lingual voicebanks, as they’re likely trained with the same targets.

Has anyone heard An Xiao singing in English? I wouldn’t mind purchasing him if he stands up a better than Ryo or Mo Chen. But if he’s still got issues, I’m afraid I’ve already got enough male background singers.

1 Like

OK, I guess I’ll answer my post, since I just picked up An Xiao.

Short version: An Xiao has the same problems with trans-lingual English as Mo Chen. There are work-arounds for those (see below).

But… he taps/lightly trills his /r/ on many of the words. There’s no fix for this.

That makes him unusable as a lead voice for me. :sob: YMMV.

Similar to Mo Chen, he’s got a lot of pronunciation issues with plosive + liquid consonants, such as /dr/ and /bl/. I can only assume there’s no native English speaker doing quality control, as it’s pretty obvious. SynthV even has special phonemes for these, which indicates awareness of the issue.

However, many of these can worked around using replacements. For example:

  • "kl" is /g l/ and with no Voicing on /g/.
  • "dr" is /jh r/
  • "tr" is /ch r/
  • "/ay/" is /ay iy/
  • "bl" is /p l/ with no Breathiness on /p/

But there’s no real workaround for the /r/. Adjusting the Strength or Duration doesn’t change how the sound is articulated, which is the actual problem.

Hopefully Dreamtonics can update their English training model to correct this.