Create new phonemes?

I want to use Synthesizer V with Scottish Gaelic lyrics. I cannot do this with an English voicebank as there are phonemes required that do not exist in English. Where do I start if I want to fix this? Thanks.

SynthesizerV doesn’t support user-created voicebanks or phonemes.


Synth v allows user edited libraires so a custom eng to scootish vb edit is possible. Just so much work would need to be done .

Your best bet it to use either Eleanor , anri, or solaris voicebanks. Than, go to libraries and start creating a scottish vb . Depending on which bank you use you can generally get away with making different language library edits .

If I knew more scottish I would work on a library edit for you . But different voices like eleanor and maki dont always translate there library edits propperly from one voice loaded to another .

I’d be happy to help however I can .

Custom dictionaries simply change the mapping of words to phonemes. They do not allow the creation of new phonemes, and are still limited by the phoneme set available for the voice library.


As of update 1.5.0 all paid AI voices can sing in English, Japanese, and Chinese, regardless of the voice’s native language.

This does require the Pro edition of SynthV Studio, but it’s possible using some combination of English and the other languages will allow you access to the phonemes you need.

It should be noted that the language setting is set per-track, so you still only have access to one phoneme set at a time. You would likely need to split the lyrics across multiple tracks based on which phonemes are needed at the time.


That’s not fully true. Making a custom library is like making a custom utau. While, the base samples are in synth v . Just like cv utaus new samples like english I and A can be created in that software . The same rules apply to synth v with different voicebanks .

Because, the english voice banks are more flexible . You can create non english samples with the phonetics . Even, for korean, chinese, japanese, irish, polish, and other languages . It’s knowing how to make the requires phonetics that limits the user and how they tune the samples .

Unless, you mean like Eleanor forte’s basic voice than, yeah there may be limitations . When, compairing her english phonetic library to her ai bank and other new english banks like anri and solaris .

But, from my perspective from doing library edits. Making a scottish libraey edit would deffiantly take a while to fine tune. But, it is possible ,

It is somewhat unclear what you mean.

If you are referring to using existing phoneme combinations to approximate a new sound, that is not the same as “creating a new phoneme”.

If you are suggesting that someone modify the voice database itself, or create an Utau voice from SynthV samples, that would have significant legal implications as both of these things violate the license agreement.

I was refferring to use pre existing samples in synth v for new sounds . But, a new library would have to be created similar to by eng to jpn library edit
Synthesizer V - Patch for English vocals download - YouTube

Unlike, my plugin that allows english synth v banks to read hirigana cv and romanji cv characters . All done by creating a new library . Creating a scottish library would be about , the same kind of work.

But, the dialect would have to have words edited . To mirror how scottish people speak and sing. So, itd be easier just to hand tune a utau/ vocaloid/ midi file in synth v instead of making a custom library .

This is not a plugin. This is a user dictionary. Like I said above, all this does is change the word-to-phoneme mapping, it does not create new phonemes or allow the voice to produce new sounds.

OP quite clearly stated that the phonemes they need do not exist in the English phoneme set.

1 Like

Edit: Sorry, I replied to @claire but it is for @Dionn_Runk

I agree with @claire - we don’t share the same terminology - you call it a custom library, patch and plugin, probably because you come from utau, but syntv call it a dictionary. We simply don’t have any possibility to create new own phonemes :frowning_face: (or what we call phonemes)

But the new multilingual voice banks are awesome and capable of many phonemes


I think there seems to be some confusion. Synt v can create nre phonetic if you use the cutom library tool. This is how I did a chinese library for eleanor and she cant read chinese text . While, saying alot of the samples wrong .

Synth v - Eng to Chinese custom library demo - YouTube

English has most phonetics you would need to create a custom language voicebank library . The issue is putting in the time to manually correct , edit, and fix any issues with the pronounciations .

As nice as the multi langual banks are . You still have to manually edit many voices that lake the multi language ability. So, Synth v would have to be treated like a vccv utau ust being converted to cv .

The best chance of making a scottish library would to edit the samples how they would be said in that language . So, unless you know a native speaker . That could be alot to take on ,

Again with the custom library you can create new phonetics by using the library tool. If you take time to see what sample is what. You can create bran new phonetics for the custom library ,

Thats why you use the custom library change how a word is phoneticlly said. If the user wants a full on scottish accented vb . She will have to spend alot of time correcting an re editing how synth v would make the voice say words .

As I pointed out in demos like this Synth v - Eng to Chinese custom library demo - YouTube

Where synth v english lacks chinese sounds . New samples can be created if people take time to see what phonetics are what in synth v . By doing so you can make new phonetics . Synth v isnt grounded to 4 characters . You can go crazy in the cutom library to make very random or very accurate samples .

The english vbs can do scottish samples . It’s just a matter of learning how the phonetics are structures in synth v .

This topic is not about singing in a Scottish accent, it is about singing in Scottish Gaelic. This is a different language with phonemes that do not exist in English.

Your “solution” does not do what you claim it does. You are not creating new phonetics or samples, you’re making a very rough approximation by using something that sounds similar to the desired outcome. This is not the same as creating a new phoneme, as shown by your own demos which are heavily accented and mispronounce almost every word.

The creation of a new phoneme would require either the original voice provider to record new source material, or extensive AI training done by Dreamtonics, or both of these things. There is no other way to enable a SynthV voice to produce an entirely new sound.

Please stop trying to mislead people.

1 Like

Again use the custom library settings and edit the phonetics . Everything you said up to this point makes no sense when synth v has a user friendly solution. But you keep ignoring it ,

It seems you’re conflating a Phonetic Dictionary with a phoneme.

A phonetic dictionary is used to convert a word in a language into phonemes. For example, the CMU Pronouncing Dictionary contains a mapping from English words into the ARPAbet phoneme set.

The ARPAbet phonemes are the same as those used by SynthV for English.

However, there are many phonemes used in various languages that are not included in the ARPAbet, such as the trilled /r/, and the Scottish /ch/, as in “loch”.

Because these phonemes don’t exist, there’s no way to make the sound of that specific phoneme in SynthV. You could use /r/ or /ch/ instead, but they are only close substitutes.

There are more complete phoneme sets, such as X-SAMPA, but many of those phonemes aren’t yet in SynthV.

Try getting a Japanese voicebank to sing in English, and you’ll see the point: there are too many phonemes in English that aren’t in Japanese to be able to pull it off. Plus, the result will be heavily accented.

That’s why the cross-lingual feature is so interesting - Dreamtonics was able train neural networks to estimate how a speaker would speak a phoneme, even though that phoneme was never part of the original set of recordings.


I think you got confuse XD When, I said about making new samples out of pre existing samples useing the custom library feature . Think of it like this japanese banks cant say the english word break . So, to create a patch in a new library setting you would set the phonetics as b r e e i k .

Becaise, you would be cross synthing samples. e i to get the english ay sound . You are creating a new sample from old samples . Heck words like or which japanese banks get wrong can have the phonetic library edit be oo a r r. You could create so many spanish samples with Eleanor’s bank for example. An she doesnt have spanish recording .

So, to get the Scootish samples in synth v. You would pretty much have to create a bran new custom voice library and build off the basic phonetics to make a scootish library edit .

I even make a synth v eleanor korean vb demo . So, its not hard to make brand new library edits for non eng / jpn voices .

No, I think you’re not seeing what people wrote.

You’re using the word “samples” wrong here.

A “sample” is a recording that’s used as the basis for synthesis.

These aren’t “samples.”

You’re talking about creating a dictionary entry for pronouncing an English word using Japanese phonetics. This isn’t a “sample”, and it’s not a “patch” - it’s just a dictionary entry.

You’re not creating anything new, and you’re not “cross-synthing” anything.

You’re using Japanese phonemes to pronounce an English word. The result sounds like an English word spoken with a Japanese accent.

Or in the case of Scottish Gaelic, the result is an Scottish Gaelic word spoken with an English accent.

So you’re method doesn’t sound like an native speaker, and it fails when there’s a word that uses a phoneme that doesn’t have an English equivalent.

The “Spanish” words won’t sound like a Spanish speaker. They’ll sound sort of like Spanish, but with an English accent.

It won’t sound “Sottish Gaelic”. It’'ll sound like Scottish-Gaelic with an English accent. And when you have a Scottish word that has a phonemes that’s not in English phoneme set, the substitution will sound very bad.


Dude most of what you said is in correct . When, I said samples I was either reffering to the samples in the vb that can be edited in the custom libraries for new sounds .
By cross synthing phontetic sounds you still get new phonetic sounds. So , yes you can get new sounds from pre existing files in any voice bank . You just cant edit the v.b. itself for bran new samples to be used as new voice samples .

Yes a custom library bank is patching . If you create the library to make the language type do things its not intended . How else do you thing library edits like my jpn edit allows eng voices to read hirigana and romanji ? It’s not a hardware patch but stuff like library edits are vocal library patches .

You dont seem to know what your talking about dude .

Everything dcuny said was correct. You have been repeatedly provided with detailed explanations about why a user dictionary does not accomplish what you claim it does.

You continue to misuse common terminology in a way that makes it clear you do not understand the topics being discussed, and it’s clear you haven’t done any research to verify your claims.

At this point you’re just being stubborn. Please stop arguing, and take some time to review the basics of digital music production before claiming that others don’t know what they are talking about.