Beta Voicebanks - Do they all work in Spanish?

Good day to the community! This is my first (selfish, interested, leechy) post.

After not using Synth-V for a while, I recently acquired Saros voicebank.

Being a native Spanish speaker I installed 1.11.0b1 and the proper beta voicebank for Saros. I am really impressed. I am not a native English speaker, but I speak it every day and I can detect many types of mispronunciations. My nice surprise is that Spanish pronunciation seems to work BETTER without any correction at all. Incredible!

Then yesterday I was aware of 1.11.0b2 released on the day before… along with ~34 new voicebank beta versions! Well, I already had Asterian and Solaria and… Bingo! They worked as promised by Eclipsed Sounds. Much earlier than I expected. And really well!

My surprise is that then I installed the rest of the beta voices for which I have license: Nine-zero, Kevin, Mai, Natalie. And they all work really well. No tweaking (well, except tick the “Eurpean z” and sometimes I prefer to have stronger oclusives (just add " ’ ") or stronger “r” (just duplicate the letter). No need to look at phonetic tables, it works much better / easier in Spanish than it did in English. I suspect that the very closed Spanish phonetic rules make the task easier.

Now: I realize there are so many NEW beta voicebanks that possibly could include the (good) Spanish capability. Many of them are Chinese or Japanese. I am REALLY interested to know if they have the functionality.

Could Dreamtonics / any user with Japanese or Chinese voices could confirm that any of these betas can sing in Spanish? Could anyone run a short test for me using Spanish lyrics? (It could be simply many “G” quarter notes singing a couple of lines of Spanish from the Internet).

Thansk a lot.

1 Like

The ones capable of Spanish cross-lingual synthesis are Saros, and any beta version in the “For 1.11.0b2” folder on the Dreamtonics download site: For 1.11.0b2 - download > English > Voice Databases > Beta Releases > For 1.11.0b2 (

(note that the individual betas are labelled as “b1” because it is the first 1.11.0 beta for that voice database, while it is the second beta for SynthV Studio itself).

The others (basically the ones developed by AHS, including the newly released Megpoid, POPY, ROSE) will have Spanish language support eventually, either in a later beta or with the final release, they just don’t have compatible versions yet.

1 Like

Thanks a lot for the quick and specific answer!

I noticed that b1 voices did work when I tested Asterian and Solaria, then I could not believe what happened with the rest of B1 I had!

Is there any consensus which voices are more likely to work better with Spanish language? The Chinese or the Japanese ones?

This isn’t from my own testing, but I think there is a generally better overlap between Japanese and Spanish pronunciation than with the other languages, so I would expect that the Japanese voices have less of an accent when singing in Spanish. Hopefully someone who has been trying it out hands-on can confirm or deny.

1 Like

Thanks a lot, I guess that someone is giving me either Yuma or Mo Chen as a Christmas present :wink:

I really appreciate what Dreamtonics is doing and also your quick answers.

Best regards

1 Like

I did this quick test with a few of the voice databases I own, including a couple Mandarin Chinese and a few Japanese options. I don’t speak Spanish, so I can’t really evaluate the quality, but Yuma seems to sound pretty good.

The SVP file used is the “Forever” demo included with Saros, and I didn’t change anything other than duplicating the main Spanish track and selecting different voices.


This is really useful! I am surprised with results. :star_struck:

First, all the voices are more than acceptable. They are very good. The worst is much better than my original expectations for Saros. They all “work” (and I’d say possibly better than some native English voices singing in English without phoneme editing).

Second, I am very surprised, the “native Japanese voices” seem to work worse than the native Chinese for Spanish :astonished:. It does not make sense to me, but I’d bet that 99% of Spanish listeners (both European and American) would say that QingSu sings in Spanish MUCH better than Yuma. I considered Mai a very good Spanish voice… and she is, but her accent is clearly Asian (which is perfect for having a variety of voices). This does not happen with QingSu!

Listening carefully all the voices and checking “who is who” I realized that Stardust is also Mandarin Chinese. But singing in Spanish her accent sounds closer to Mai or Yuma. I don’t know… perhaps what I hear is not exactly accent (phonemes) but the way to use the voice (more “nasal” longer time), and Qing Su sounds better. Difficult to say, listening carefully QingSu is louder, this tends to “sound better”, but there are definitely more “Asian style or Asian,… something” in the other voices.

Anyway I am SOOOO surprised with the results! The worst one is good, better than I could have imagined. :clap: :clap: :clap: :clap:

I would suggest a minor correction for the European pronunciation, it is a minor thing. It is the “Z” sound when it is at the end of the syllable. ZA, CE, CI, ZO, ZU lyrics are all properly vocalized with phoneme “s” or “C” (depending on American / European), but in words like “FELIZ” the phoneme is always translated to “s” instead of “C”. I had to edit manually for a more natural sound. Being final, also I had to lengthen and strengthen this phoneme “C” to work. It works 100% with manual editing (and not editing does not sound bad for European Spanish, and it is 100% perfect for American). But it is the ONLY minor accent issue I detected.

I think think that phonetics are easier in Spanish than in English. To begin with we don’t have words that are written the same but pronounced differently depending on the meening (like “wind” or “lead”), and our 5 vowel characters correspond to just 5 sounds, not 10 or more. So I was expecting that Spanish could work better… BUT NOT IN THE FIRST BETAS! Congratulations! :clap: :clap: :clap: :clap: :clap: :clap:

Ah! Thanks for putting together this demo. I think it will pay off in sales.


I have detected some bugs. Like when writing some words, a phoneme is sometimes not generated for one of the letters. Weirdly enough, this happems randomly, as the the same word in another line is detected correctly.
The sound I am struggling to get a correct pronunciation is the “J” (Ja, Je/Ge, Ji/Ji, Jo, Ju). In the song I was working, it still sounded like an aspired H.
Might be the voice banks or the crosdlingual synthesis. Also you can still tell the current VB have a bit of an accent.
But yeah, all in all,is a great improvement from my previous attempt to generate Spanish by mixing English and Japanese. I guess some refining and a couple of native Speaking voice banks (at least a male and a female) is all is needed now.

1 Like

I can hear what you say, I did not notice. And I agree but I do not think it is very bad. I tried for a few minutes and I noticed:

  1. It helps to lengthen / strengthen the phoneme of the “j” letter (phoneme " x "). Set both duration and strength to max. and it will sound like average Spanish or even “too much”.

  2. Not enough? Depending on context (especially on previous note), you can try to force a longer “vocal mute” during the consonant to make it stronger. By adding " ’ " in the lyrics (no need to even edit phoneme). I mean, write " 'JA 'JE 'JI 'JO 'JU ". This makes some weak consonants (for Spanish standards) a lot stronger. It works VERY well with letter “T” (so soft that it sounds like a “D”). The good thing is that you can type all these corrections intuitively, no need to look at phoneme charts like I did when I struggled with ENG-JAP-SPA dictionaries. It is easy to exaggerate these sounds to imitate a strong Spanish accent in my opinion.

Also, to get a good initial “R” (like Ratón) you need to do something. Imagine the simplest option and… Bingo, this is right! Write RRatón. It sounds really natural. For best results, sing " rra 'ton "

Today I used Synth-V to “sing” Happy Birthday to a friend before office work. Final “z” in “FELIZ” required manual editing. Be aware if you wanna sing “Feliz Navidad”!

Well, I must write something, even if it is some test for pure fun. I would like other Spanish language users to post here (or wherever is best) to share experiences and help each other. At the end of the day this technology is brand new for our langauge. I am not sure if this is best done in English or Spanish, I can see advantages and disadvantages.

Thanks a lot for your suggestions!
the extra R for that starting RR sound I did try that just out of intuition and it did work, indeed!!
The other tricks you suggest I was not aware of… I will try them as soon as I can.
Thanks a lot for the tips again!!

1 Like

Regarding the rr phoneme, Eclipsed Sounds also says to make sure you adjust the pitch curve accordingly (ie set the note to Manual pitch mode and draw a quick vibrato-type pattern with the mouse).

Spanish Feedback | Eclipsed Sounds

Note: The rr phoneme requires a shaky pitch curve in order to sound correct, which is not always correctly generated in the current beta version. Please be sure to add this “vibrato” pitch over the phoneme for the correct sound, or use AI pitch retakes until it sounds correct. We will be focusing on improving this over time, but it should naturally improve as Synthesizer V Studio receives more Spanish language data in the future.


Not really!!!

This patchy solution was required in beta-1, it is artificial, complex and I did not like the results.

Writing “RR” (as we normally do in the middle of a word) makes the distinction between the 2 phonemes… exactly in the way they are written in the middle of a word. This by itself is a giant step in the evolution of Synth-V Spanish phonetics. One of the toughest. And so EASY to do! This happened 3 days ago in the second beta, not before.

At the beginning of a Spanish word, grapheme “R” is pronounced always as “RR” (Synth-V does not know… yet) but the fix is what any little child could do intuitively: Write “RR” (as in the middle of a word). And it works. No need to edit pitch curve or anything, it just WORKS. And really well.

If we use single “R” at the beginning of a word, this does not exist in Spanish, but sounds exactly as many foreigners (without this sound in their native language) try to pronouce “RR”,

As I mentioned before, I had to constantly look at phonetics tables (arpabet?) to guide Synth-V to sing proper English. No need so far in Spanish, I just tried simple intuitive solutions and they simply WORK. Magic. This did not happen in the first beta (Saros only). Congrats to the team.


I spent like 2 hours with Synth-V in the last 4 or 6 months, so I am no expert at all. Iearnt many tricks months ago… and now they are all useless. Things are now way easier. I am sure that any Spanish user spending a few days will go much further than me.

1 Like