I'd like to hear examples of what other users have accomplished with Synsy. I feel that Peter implemented the feature, at least in part, to give voice to my music. I'd like to show my appreciation by singing a song to/for Peter.

Donny

Excellent idea Don, and absolutely very commendable.

My results with Synsy have been less than desirable, partly possibly due to the Japanese sounding phonemes that have been used.

More information on Synsy is available here

Of course, YMMV

Fellow forum fember David Cuny has also been developing SynSinger, a similar product and one you might explore also.

Ian,

Yes, I've been working with David Cuny.

Voice synthesis, especially singing, is in it's infancy. I'm hoping to get David connected to the engineering staff at Dynavox-Tobii who have developed the most realistic-sounding speaking voice that I've ever heard. It seems to me that singing voices would be a logical progression. I don't know if David was successful in contacting their R&D engineers.

Dynavox was recently acquired by a Japanese company, Tobii, so I'm anxious to see/hear what they do with the technology.

Another interesting technology is a voice bank (I've forgotten the name) where they take snippets of volunteer's speech then concatenate them and process the results through a Melodyn-type processor. That idea shows a great potential.

Donny

I've done a couple songs using Sinsy - for example, Words Like Smoke.

The early BiaB support for Sinsy had a number of problems which might have turned people off from using it. For example, syllables had short rests between them, so words often were mispronounced. There was a bug which caused some notes to be exported as naturals instead of flats. I think these have both been corrected, although I've never gotten confirmation back from my bug reports.

I wish that the people working on Sinsy had used a native English speaker, because the /R/ and /L/ aren't very distinct. But the voices are very natural, and there are a number of YouTube videos showing off what's possible.

Anyone who's familiar with voice synthesis won't have trouble finding out about Sinsy - there's a ton of information on the internet. Basically, it uses HMM (Hidden Markov Model) synthesis and then modifies it from spoken to sung speech.

The source for Sinsy has been released on SourceForge. Unfortunately, it's beyond my skill to build it, and they apparently only included a single Japanese voice.


I've heard demos that have spliced together segments of actual speech phonemes along with formant synthesized speech. The results take on a strong character of the original speaker.

Edit: I see on the Sinsy SourceForge discussion board from a couple weeks ago that they're planning on releasing the English version in the near future. Part of the issue seems to be that the English version was done in an ad-hoc manner.

They suggest that if you can't wait that long, you could do it yourself by modifying the source code.

David,

I believe the name of the project is VocalID (qv).

They are still in the process of gathering voice samples.

Donny

Don,

I believe you maybe thinking of ++ Vocal ID ++. Another possibility is ++ Model Talker ++

Charlie,

Yes, Vocal ID is the one I had in mind.

I am an honorary member but I've fallen behind in their progress. They have been working primarily on speech and not singing which seems a logical progression.

Thanks,

Donny

Hi Don, I have been working with Synsy for the last couple of days and my problem is that the entire vocal cuts out except for a faint "tsk" sound after every 5 or 6 bars. The phrases that do sound are quite clear although there is quite an "accent" to the words. I've downloaded a new version several times and it continues to cut out every few bars. I'm beginning to wonder if it is a problem with the track itself. I checked the Audio Edit and there are spaces where it shows almost no activity and then it will pick up again several bars later. Can't imagine what is causing this. Would you or any of your friends on here care to venture a guess of why this is happening or how to fix it?

Wonderful to hear from you, my friend.

Are you losing your beautiful voice so starting reliance on voice synthesis?

Singing voice as opposed to speaking voice is another world entirely. I just received a new Dynavox 15" Speech Synthesis Device that speaks perfectly. The problem is in making it sing.

I would really like to team David Cuny up with the R&D Engineers at Dynavox-Tobii.

I have an older Dynavox VMax (qv) that you are welcome to have for conversation purposes. Just nod your head and I'll ship it out.

You may enjoy learning that Illinois State University's Dr Kim McCord has honored me by establishing The Don Gaynor Special Needs Music Department. I'm thrilled. Unfortunately, Professor McCord is being forced into early retirement due to serious family health issues and the University has no plans to continue her program, the top-rated such program in the nation.

Again, it's awesome hearing from you.

Appropriate affection.

Donny

Sorry - I thought I was in PM mode. ^^^

Yes, Don, I really have no voice quality anymore. I believe it may be a thyroid issue. No trouble speaking just the hoarseness that keeps me from a good vocal quality.
I am finding it more and more difficult to learn to use these new technical advances so though I appreciate your offer of the Dynavox, I fear that it wouldn't be much use to me and perhaps someone else could make better use of it than I. I haven't even mastered all the things available to me in BIAB. There is just so much to learn and absorb so it can be put to use but I'm working on it as time allows.

I'd be happy to have a look at it. Could you send the Sound.XML file that BiaB generates? If you installed BiaB on your C: drive, You should be able to find the file here:

C:\bb\data\SRequest\Sound.XML

If not, you can find it by going into BIAB, choosing Lyrics from the menu, and selecting Vocal Synth-Manual Mode... from the dropdown.

BiaB will display a dialog box with the following message:

C:\bb\data\SRequest\Sound.XML is now generated

This indicates where the file is located.

My own voice synthesis program has a few timing bugs I'm still struggling to track down. The voice quality is OK, but it's hard to understand at fast tempos, and is a lot more synthetic than Sinsy.

But if you're interested in trying it out, let me know. The primary voice is male, but I've got the option for a female voice. Here's an example of what it sounds like: synSinger Female Voice Demo. Obviously, the female voice needs more work.

Val,

Actually, Dynavox is not very technical - it speaks what you type into it's touch-screen keyboard. Common phrases are pre-programmed and assigned to hot keys. Literally hundreds of common phrases make typing virtually unnecessary.

Also, it's not a toy. It's a full blown Windows 7 computer with WIFI and a fairly decent audio amplifier.

I would be happy to give it to you. It's just gathering dust because I have three. I will have to locate the charger/adapter though.

I'm told that my new 15" cost $15K so the VA is taking excellent care of your old Irish friend.

Donny

Hi Dcuny
Oh I tried it a couple of times but I didnt think you could get it as good as that, words like smoke is really nice song.If you take away the japanese sound its very good.
Hugh

Thanks for responding to my post and for your offer to take a look at my problem. I have just spent the whole day re-downloading all my BIAB files to the C drive, hoping this may make a difference. I had been using the hard drive version where I first experienced the problem so I dug out my back up copy and tried it on that, thinking perhaps some files had become corrupted on the original hard drive version. But It did the same thing with my back up copy so I have been re-downloading everything and putting it in my C drive and when I get that finished, I'll try Synsy again. If that hasn't solved the problem, I will get back to you and send that file that you needed. Thanks again for offering to help.
I did go and listen to the female voice in your It's not bad but I use a male vocal for a lot of my songs so I would be more interested in hearing the male voice. I'm wondering if maybe some malware got into my computer when last week one of those scams for computer repair came up on my screen and locked my computer and I had to just shut it down from the task manager. It messed with the files in the boot sector and I had to use the restore and repair to get it back up and running.

Don, I want to thank you again for your offer of the Dynavox. I became interested when I learned that it was Windows 7, because that is what I'm familiar with. I tried Windows 10 and did not like it at all. If you can wait for about 3 months before you send it I will then have a place to put it where I could use it. We hope to add another room onto our apartment this spring. Hubby has been drawing up plans because we are just so "stuffed" into our tiny apartment since we converted my (computer room, music room, office) into a room for our little great grand daughter and moved everything into our already overflowing bedroom. One can hardly move around to find everything. That will give you some time to find the adapter/charger too! Thanks again!
____Val

Here's a link to the male voice: Twinkle, Twinkle, Little Star #16.

Dear Val,

My brief description doesn't do the Dynavox V-Max justice. It can be configured for hands-free Eye Tracking (with a separate accessory). It has WIFI for built-in internet access. I will also include a WIFI Range Extender so you can use it outdoors (within line of sight). It has a scratch resistant touch screen and 2 extra USB Ports for other accessories such as a WIFI or Bluetooth wireless keyboard.

Here is a picture and complete details.

http://www.dynavoxtech.com/products/vmaxplus/specs/

There is an 1800 page Owners' Manual pre-loaded on it's hard drive.

Don

PS: I should mention - It's CPU (Intel Adam) is NOT sufficiently robust to run BIAB.

Thanks Don for sending that link! I've been checking it out and I see that there are two ways to order it so that you can do more than just the "speech" on it. So is this one standard or dedicated? I'm thinking that the standard would allow for a game of Solitaire now and then, Am I correct? I think I should send you a request on Facebook to continue our conversations. OK? Been working on downloading BIAB files to my C: drive so then I can try the Sensy again. I discovered today that my computer had been invaded by malicious code that was not caught by my Anti-virus so I changed my antivirus to Kaspersky today and it immediately caught it in the first 5 minutes of the scan! Now I can go back to work.

Val

Yes, it allows access to it's internal Windows 7 OS.

Agreed, we have gotten way off topic so let's move it to Facebook or PMs.

Donny

BTW, it has a few games pre-installed and you can download Solitaire from the web. I love Hangaroo (like Hangman). It has a built in camera and will clone any remote control so you'll have control of TV, etc.

It has a nice-sounding mp3 player and room for thousands of mp3s. You can watch videos from YT or add a USB Video Player. The screen is crystal clear.

FYI - there's another vocal synthesis program that's being developed called Moresampler. It's specifically targeted at singing synthesis.

I started working on another song (using synSinger) after listening to Don's song based on Yesterday's chord progression.

Unfortunately, BiaB has a lot of "undocumented features" when it comes to the lyric mode.

Here's a fun use case of a dizzying, but (mostly) benign bug:

1. Create a new song.
2. Enter notation mode.
3. Choose Editable Notation Mode.
4. Click in the first bar to enter a note.
5. Scroll to bar 5.
6. Click on bar 5 to enter a note.
7. Click L to enter Note based lyrics.
8. Click L to exit Note based lyrics.
9. Click L to enter Note based lyrics.

When you click the L to enter the mode, several (wrong) things happen:

• The screen draws bars #1-4. BUG
• The screen draws bars #5-8.
• The note on bar #5 is drawn red, to show its active.
• The note is then re-drawn in black. BUG

It's a bit dizzying, but apparently benign.

But after working on the second half of the song verse, I found that the lyrics to the first half of the verse had become corrupted. (This has happened on multiple occasions, but I've yet to find a use case to duplicate it).

So I deleted all the lyrics (not the notes) using the Edit Lyrics dialog. I then pressed the Play button, at which point all the notes vanished.

I've send in a bug report on that one.

Also, has anyone noticed in the Copy Special dialog there's on option Copy 1st Chorus to whole song that's set on by default? I sent in a bug report on this as well.

Anyway, as you might have guessed, what started as a "fun" project to play with vocal synthesis in BiaB turned into a complete waste of time (except for filling out bug reports).

David,

Moresampler sounds great except it's sung in Japanese so it's impossible to evaluate it's handling of English.

The Japanese developer sounds like you. Another David vs Goliath.

I imagine that if you were to team with Dynavox, you would be required to sign a noncomp and forfeit your work.

Maybe consider partnering with a college or university near you.

Donny

Sorry, try this link for some examples in English.

David,

To these tired old ears it sounds staccato and truncated. Each syllable is chopped short, no decay.

Donny

Very interesting Sinsy. I tried to play a bit with it and the quality is very high.

Unfortunately when I tried to make it singing very low notes (in english) the result is not good at all.

The reason why I'm interested in low notes is that I would like to create a solfege midi bank for this project

https://github.com/wcgbg/solfege-samples

If you check it, you will see there is already one synthetic voice bank that's done with Alter/Ego. I would like to add synthetic voices. Do you know any other software (free sw) that I could use? Ahem.... I see you're developing one (I read your blog). What about your software?
And do you have any idea how I can improve the quality of very low singing with Sinsy?

Thanks a lot David.
Gianmaria, Geneva

Hello David,

I found this post very interesting. I didn't know sinsy and your (David) blog site.

I tried a bit to use sinsy for a specific task. I would like to create a soundfont midi bank containing solfege note (do, re, mi etc.). Sinsy pronunciation is a bit japanese... but that's ok. Something less acceptable (for what I need) is that when ask sinsy to sing low note it generates very robotic voice. Maybe I'm doing something wrong... ? Do you know if there is any way to fix this type of issue?

Here you can find the project I'm trying to collaborate:

https://github.com/wcgbg/solfege-samples

I also listened the result of your program and would be great for me to try to generate the notes (do, re etc.) with your program. Is that possible?

Thank you, Gianmaria
Geneva

Yep. I've gotten a hold of you via email.

Interesting thread but one point was made with which I disagree heartily. This is not in it's infancy. Bob Moog built a Vocoder, which this really is but under a souped up 2000s name, in 1968. That makes it 50 year old technology. It was used in the soundtrack for A Clockwork Orange, and on a LOT of songs. Kraftwerk used it all over Autobahn, and Jeff Lynn used it in ELO's Mr Blue Skies in 1976. Find a copy of Kenny Loggins Keep The Fire and it opens the song. Asia used it during live shows because nobody could really sing harmony well. Geoff Downes sang into the triggering mic and played the parts.

I had the Digitech version of these things for a long time. The old school units were very primitive and the voices sounded quite robotic, nowhere near as real as the voice processing tools work today, but that is normal product evolution that it get better every few years. The Roland Boss VE-20 is outstanding. And just $280. I have seen a lot of bands use that rather than truth the drummer to play with a click track so when the chorus hits and the BGV kick in they are in time. If I was still playing I'd buy the Roland.

Hi, Eddie.

Vocoding takes a "real" voice and uses it to modulate audio. It's a transformation of the original sound.

Voice synthesis is creating a new vocal from scratch, where you can enter lyrics and pitch information, and the computer generates the vocal without human intervention.

Computers driving a vocoder are one way of generating that sound.

Greg Mackie, an audio genius by anyone's estimation, spent a lot of time on a vocal synthesis circuit and eventually gave up after realizing that technology wasn't there YET!


I suggested that David contact the R&D folks at Dynavox. They have the most realistic-sounding spoken voice that I've ever heard.