Synthesizer V - Singing voice synth for Win/Linux/Mac (Free & Payware)

[ZincT]

@Pelicanomicon

One thing I am finding with Ryo is that he has quite an accent (Japanese?) in places on certain words. I was kind of surprised by that as I thought the multi-lingual support overcame such things, and, as far as I can tell, I have the English dictionary selected. I guess he is at v1.00 so maybe they will improve this as time goes by. Now I'm thinking that I should have bought Kevin who I assume doesn't have this issue? Maybe there's something I need to change in the configuration?

Any ideas?

Thank you.

[Pelicanomicon]

One thing I am finding with Ryo is that he has quite an accent (Japanese?) in places on certain words.

Ryo is Japanese

I was kind of surprised by that as I thought the multi-lingual support overcame such things, and, as far as I can tell, I have the English dictionary selected. I guess he is at v1.00 so maybe they will improve this as time goes by. Now I'm thinking that I should have bought Kevin who I assume doesn't have this issue? Maybe there's something I need to change in the configuration?
Any ideas?
Thank you.

Kevin's pronunciation is definitely "American Englisher" than Ryo's, yes, and he sounds good out of the box. The short answer: I'm afraid there isn't a general configuration for Ryo. Until Dreamtonics improves the Cross-Lingual modeling (personally, I believe this will happen at some point), his mannerisms are handled by tuning or tools like Melodyne.

But if you prefer Ryo's tone and don't mind some trial and error, here are some tips:

Tempo, note length and phoneme context affect pronunciation, so what works in one song will sound weird in another.

1. Probably the easiest thing you could try is creative substitution of phonemes. I did this A LOT with KAITO V3 English, sitting at my desk and sounding out lyrics in the slowest, doofiest way to puzzle out phones. Like Dory!
   
   Ex.: The SVS default pronunciation for "time" is /t ay m/ .
   For Ryo, this can sound more like /t ah eh m/
   As a workaround, you could stick a /y/ in there and turn the Duration and Strength of it way down in the Note Properties | Timing and Phonemes panel: /t ay y m/ .
2. You can duplicate certain phonemes and SVS will enunciate them more clearly instead of repeating them.
   Ex.: Default "brother" = /b r ah/ /dh er/
   Ryo workaround = /b r r ah/ /dh er er/
   Going back to Tip #1, also /b er ah/ /dh er er/ , /b ax r ah/ /dh er er/, etc.
3. You can stick a phoneme on its own note and tinker with Note Properties and parameter lanes. But you'll have to adjust the note length of the other parts of the word as well to get the whole word to sound natural.
   Ex.: Default "alright" = /ao/ /l r ay t/
   Ryo workaround = /ao l/ /l/ /r ay/ /t/
4. You can use the "Voicing" parameter lane to simulate tongue position/mouth openness. Useful for Ls when you want it to sound like the tongue is pressing against the palate, or back-of-mouth Rs. Draw dips and curves where those sounds fall. BTW, the only consonant a Japanese word can ever end in is "n", so Voicing can help with some of the unwanted "openness"/"pureness" of non-native English voicebanks. I used the equivalent parameter in VOCALOID to reduce English uncanniness--nobody sings an "m" with their mouth wide open!
5. Finally, you can use the "Gender" parameter lane for formant shaping. SVS responds to Gender tweaking dramatically, so it might be better to start with small, gentle curves.

That's what I can think of off the top of my head. I hope you don't regret your purchase too much, @ZincT

[ZincT]

Thanks @Pelicanomicon

I really appreciate your detailed reply. Phonemes seem to be the way to go to get better control over pronunciation. I also probably need to go through a few of the video tutorials again, now that I have VS Pro and full voicebanks.

I was also pleasantly surprised at the tone variation available for Ryo using the Gender slider, so I don't regret buying Ryo as I do like his tone but I will probably buy Kevin as well

Do you know if Eclipsed Sounds are working on any male vb's, as I find Solaria works pretty much how I want it to out-of-the-box. I have also heard some great work using Eleanor Forte AI so will probably add her to the collection at some point. I expect by the end of 22 we will be spoilt for choice voicebank-wise.

[Pelicanomicon]

Oh man, ppl were asking Eclipsed Sounds for a male voice from the start of the crowdfund! They had a livestream on 21 May 2021 where they responded at around 21:00. It's been blocked from embeds, but you can find it on their YouTube channel.

We can't answer this completely because it all depends on how well SOLARIS does. Right? So, like we would love to. We love Synthesizer V, we love English vocals for Synthesizer V. We, like...That would be amazing if we could do that. Right now, we just need to make sure. That won't be possible unless SOLARIS does well. So our future plans are to expand--and this is on our crowdfund and this is on our company website. Like, our company plan on the whole is to expand English vocal synthesis. And we would love to make another one. But it really depends on how well SOLARIS does. So we'll see!

I wonder if Emma Rowley has any friends in Nashville she could convince...

[Pelicanomicon]

Some related news (translations by DeepL)

Dreamtonics has revealed its high-quality AI TTS program teased last October: "VOICEPEAK" for Windows/Mac/Linux. What distinguishes this from TTS programs popular with Japanese content creators is that Dreamtonics/AHS are explicitly allowing businesses to commercially use the exported audio.

The flagship product is titled 「VOICEPEAK 商用可能 6ナレーターセット」 (VOICESPEAK For Business 6-Narrator Set). It comes with 3 Adult Females and 3 Adult Males, plus 1 Female Child.

Product Page:
https://www.ah-soft.com/voice/6nare/

個人ユーザー様のご利用はもちろん、教育機関の方や法人様など、様々な商用・業務用途でのご利用も可能です。
It can be used by individual users as well as by educational institutions and corporations for various commercial and business purposes.

VOICEPEAKは、Dreamtonics社が開発した高品質なAI音声合成エンジン「Syllaflow」を搭載。
人の発声と区別のつかないほど、より高次元な読み上げが可能です。
また、感情表現にも優れており、喜怒哀楽などの感情を込めた読み上げも自然に行います。
VOICEPEAK is equipped with Syllaflow, a high quality AI speech synthesis engine developed by Dreamtonics. Syllaflow is a high-quality AI voice synthesis engine developed by Dreamtonics. Syllaflow also excels in expressing emotions, and can naturally read out words with emotions such as joy, anger, sadness, and sorrow.

English EULA: https://www.ah-soft.com/voice/6nare/eula_e.html

Like Synthesizer V Studio, Cross-Lingual Synthesis is planned for VOICEPEAK. This is mentioned in a new DTM Station interview with AHS president Ogata Tomohide 「音声合成業界に激震! もはや人間の喋り声、入力文字読み上げソフトVOICEPEAKはビジネス用途でも自由に利用可能」 ("Shaking Up The Voice Synthesis Industry! Incredibly Human Text-To-Speech Software VOICEPEAK Can Be Freely Used For Business"). (Gotta love these DTMS headlines, they're so energetic...like Super Sentai episode titles!)
https://www.dtmstation.com/archives/55090.html

－－このVOICEPEAKで喋ってくれるのは日本語のみのようですが、今後はどうなるのでしょうか？
尾形：Synhesizer Vと同じような展開にしていく予定です。つまり英語、中国語を喋れるようにするだけでなく、日本人の声のデータベースを使って英語や中国語を喋ったり、逆に英語圏の人の声を元にしたデータベースで日本語を喋らせる……ということが可能になると思います。
--It seems that this VOICEPEAK only speaks in Japanese, but what will happen in the future?
Ogata: We are planning to develop it in the same way as Synthesizer V. In other words, in addition to being able to speak English and Chinese, it will be possible to use a database of Japanese voices to speak English and Chinese, and conversely, to use a database based on the voices of English-speaking people to speak Japanese.

There's a free trial, but it only speaks Japanese. However, it does recognize installation language and you can change the Japanese UI to English--a good sign. Sound quality really is clear.

If you're feelin' risky, you can digitally preorder from AHS for the promotional price of ¥15,800 until 30 April 2022 23:50 ($138.15 USD in cart atm). After that, it goes up to ¥23,800. Click the blue DL予約 ("Preorder") button to put it in the region-aware Cleverbridge cart.
Store Page: https://www.ah-soft.com/product/download.html#voice

[ZincT]

@Pelicanomicon

Voicepeak sounds very realistic. I guess it's a natural progression of their SV technology.

Incidentally, after spending some time with Ryo tweaking phonemes etc, I decided to contact Dreamtonics and see if I could swap him for Kevin.

They were really helpful and quick to respond. Also, so that I could get the special offer again (VS Pro plus one VB), they actually refunded the whole order. I then re-ordered it with Kevin rather than Ryo. I have tried out Kevin in a couple of projects and I think he sounds much more in line with my needs.

So hats off to Dreamtonics for their customer service

[ZincT]

Just for fun here's a work-in-progress Solaria and Kevin doing their best Linda and Aaron impression I'm using it to get more practice working with both voicebanks.

[Spitfire31]

Just for fun here's a work-in-progress Solaria and Kevin doing their best Linda and Aaron impression I'm using it to get more practice working with both voicebanks.

Excellent 'work-in-progress'! Both voices sound really convincing – I never get that old uncanny valley feeling that I'm hearing programmed digital singers. And no need whatsoever for cheating subtitling. These are good times!

[ZincT]

Just for fun here's a work-in-progress Solaria and Kevin doing their best Linda and Aaron impression I'm using it to get more practice working with both voicebanks.

Excellent 'work-in-progress'! Both voices sound really convincing – I never get that old uncanny valley feeling that I'm hearing programmed digital singers. And no need whatsoever for cheating subtitling. These are good times!

Thank you for the feedback @Spitfire31. Yes, it's amazing what can be done in 2022 with VS digital singers!

I tried to be faithful to the original song but there are some "vocal riffs" that Aaron does which I cannot get Kevin to do (maybe my inexperience with VS pro or maybe it's to be expected given Aaron's vocal talent).

Solaria remains my favourite though and so easy to get great results.

Kevin seems to have a lot of sibilance like on the "ch" of much which even Fabfilter DS struggled to remove so I had to dip the loudness on those parts. On real vocals I often use Melodyne to edit sibilance but I wanted to see what I could achieve as much as possible in the VS Pro plugin and without pre-rendering the vocals.

[Bojo Software]

Just for fun here's a work-in-progress Solaria and Kevin doing their best Linda and Aaron impression I'm using it to get more practice working with both voicebanks.

It sounds great. I find Kevin much harder to use than Solaria, but here he sounds great.

[Pelicanomicon]

Just for fun here's a work-in-progress Solaria and Kevin doing their best Linda and Aaron impression I'm using it to get more practice working with both voicebanks.

Solaria sounds great as always, but I really like your mature, husky Kevin @ZincT Good tuning!

[ZincT]

It sounds great. I find Kevin much harder to use than Solaria, but here he sounds great.

Thanks @Bojo Software The auto-tuning seems to create much more random results with Kevin than with Solaria. Maybe it will improve when they update it.

Solaria sounds great as always, but I really like your mature, husky Kevin @ZincT Good tuning!

Thanks @Pelicanomicon. I altered Kevin's tension and gender a little to get this result. I also discovered that VS Synth Pro can import lyrics and melodies from "karaoke" MIDI files, although the lyrics aren't 100% assigned to correct notes. Presumably it can't automatically assign multi syllabic words to notes as it doesn't know how the words are sung. Pretty useful feature though.

[Pelicanomicon]

News for Mac users!

https://twitter.com/dreamtonics_en/stat ... 5959217154

Synthesizer V Studio now runs natively on Apple Silicon! Get the version 1.5.1 update to get a 300% rendering speed boost on M1 series chips. For Pro version users, native Apple Silicon support also extends to VST3 and AU plugins.

* This is a macOS-only update.

https://twitter.com/dreamtonics_en/stat ... 9080033283

If you have purchased Synthesizer V Studio Pro, please enter your activation code here to get the latest installer.
https://auth.dreamtonics.com/store/download

The above link lets customers of the Dreamtonics store download installers for their purchases by entering their activation codes, so that's handy.

[Bojo Software]

I've finally made a song with the Kevin voice. It's a cover of Low Roars Give Up.

[ZincT]

@Bojo Software

Excellent cover. I didn't know the song but had a listen and yours sounds great. The Solaria harmony was a nice touch.
I also really enjoyed the backing track. What did you use for the instrumentation?