Forum

Please consider registering
guest

Log In

Lost password?
Advanced Search

— Forum Scope —




— Match —





— Forum Options —





Minimum search word length is 4 characters - maximum search word length is 84 characters

The forums are currently locked and only available for read only access
This topic is locked No permission to create posts
143 Registered Users . . .
Apr 7 2004
10:23
quetzalcoatl
Member
Members
Forum Posts: 173
Member Since:
Feb 26 2004
Offline

We have 143 people who’ve registered at this forum, and around 20 people who are active posters in the forum, what’s going on?

If you have registered here please post and help make this a vibrant community :D

Apr 7 2004
22:09
rickpaul
Member
Members
Forum Posts: 18
Member Since:
Apr 7 2004
Offline

Hi, I’m brand new to this forum, and maybe user 144 since I just registered after reading this message (and a bunch of others earlier).

I’m pretty new to LOLA, and Vocaloid in general, and must say its been challenging trying to get useful results. I’m currently using LOLA with SONAR Producer 3.1.1 via ReWire.

I have not been able to get native English sounding pronunciations out of “her” — I’ve seen some others mention her sounding Japanese, but I’d more of guessed Bosnian or something Eastern European. I’m wondering how much can be done on this front by tweaking the phonemes? Just as an example, the song I’m working on has the word “caught” in it, which Vocaloid makes “kh Q t”, which would suggest to me that it should be pronounced like “cot” based on the pronunciation guide. However, to me, it sounds more like a slowed down version of “coat” to emphasize the dipthong of the long “o” sound (i.e. KOH-oot), where what I’m looking for is more like “KAWT”.

I’m also just starting to experiment with the various articulations and control envelopes to try and humanize the sound, but it is SO slow to do because every time I want to play back it takes something like 7 minutes to render the song, even if I’ve only changed one word on each of the three tracks (I’m trying to use it for three part background vocals on a roughly 3:20 pop song). I suppose I could just try and recreate something similar in a separate standalone instance with only one short segment, but all the double tweaking would be frustrating, too.

Anyway, I wanted to honor the request to check in, and, if anyone has any good advice on the pronunciations or humanizing, I’d definitely appreciate it.

Rick

Rick Paul
Closet Cowboy Music
[url:1jjsmswa]http://home.earthlink.net[/url:1jjsmswa]
[url:1jjsmswa]http://www.soundclick.com/rickpaul[/url:1jjsmswa]

Apr 7 2004
23:26
gray
Member
Members
Forum Posts: 304
Member Since:
Feb 27 2004
Offline

Try O: instead of the Q. The handbook shows O: in the word taught in the examples. Try breaking your melody in to small parts for faster speed in assembling a song. Make part 1 very small for experimenting to get the parameters sounding their best. Look at some of the midi files posted here to get a better idea of how to get decent sounds, however dont put that in stone as everyone here is still experimenting. What works for 1 song may not for the next. Try to pick songs with pitch range suited to vocaloid till someone here figures out how to make it sound good in other ranges. When you do a song post it in midi format so others can look at what you did. You may not know squat but do something that others haven’t thought of yet. And if you do something right that I aint doing, I’m gonna find it I betcha.

Apr 8 2004
00:39
quetzalcoatl
Member
Members
Forum Posts: 173
Member Since:
Feb 26 2004
Offline

Hi rickpaul, welcome dude. You’ve kind of got to think laterally with Vocaloid to force it to sing what you want. You have to think of other words that [i:1b9emi47]sound like[/i:1b9emi47].. For “caught” I would try “caht”, and I’m assuming you want an American accent, which to an English ear sounds more like c-ah-t, the english pronunciation being more like “court”, silent “r” not rolling “r”, and not like “fought” which in American sounds like “fart” oops LOL <img class=” /> ..well I think you get the picture anyway. But I’ll bet that accents are playing a big part in the frustrations with Vocaloid. We’ve decided that Lola is in fact a Taiwanese Londoner, so she only understands “One number 18, two number 27 and fwee egg-flied lice”, but we’re teaching her English and she’s coming along nicely <img class=” />

Apr 8 2004
16:11
rickpaul
Member
Members
Forum Posts: 18
Member Since:
Apr 7 2004
Offline

Thanks, Quetzalcoatl and gray, for the ideas on phonemes and accents and such. I had noticed the “taught” example with “O:” in the phoneme list and wondered why it would treat “caught” differently than “taught”. However, I just tried a quick experiment with that, and it doesn’t seem any better. In fact, I’m not sure I can tell the difference between that and the “Q”. I also tried just the “O” as suggested, but then somehow the “kh” sound on the front of the word gets dropped, so it is more like “ought” (or maybe a bit more like “oat” or think of a German “o” plus umlaut.

I also appreciate the tip on posting .MID file examples. Thus far, I don’t think I’m doing anything interesting — I haven’t even added attacks and such, and have only minorly modified a few of the phonemes.

I did also look at some of the files posted by others, especially some of those that felt the most realistic to me, and have taken a few ideas for when I get back to adding humanizing elements (have to take a break first to work on other aspects of the arrangement of the song as I really want to get a rough mix finished by the end of the week, and even using Lola’s default pronunciations and no inflections may work okay in the background for the rough mix purposes). However, even with the best of the recordings I heard, it still felt to me like Lola sounded like she had a foreign (and I don’t just mean UK English) accent, and still sounded more like something mechanical than human, or at least beyond-Cher-“Believe” processing of a real voice.

One thing I did notice was that none of the examples I looked at were using the resonance controls, and I was wondering if those might help in any of this in that I do know, with real singing, how you use the various sources of resonance in your body can make a big difference in the overall tone, and that evolves during the course of the performance. Not even sure where to start experimenting with that stuff, though, as there are so many controls in that area, and I have no clue which of the resonators 1-4 are meant to simulate, for example, nasal cavities, chest, etc., if indeed they are meant to correspond to physical aspects in that way.

Rick

Apr 8 2004
16:18
andromeda
Member
Members
Forum Posts: 188
Member Since:
Feb 27 2004
Offline

I too would like to know all these things – we are still waiting for more tutorials. Sadly, nothing yet.
By the way, the vowel sounds also change with pitch, as with real singers…
It makes it very difficult to to get things right, especially when the rendering time takes so long and you get fed up waiting.
Cheers
Chris

Apr 8 2004
21:32
quetzalcoatl
Member
Members
Forum Posts: 173
Member Since:
Feb 26 2004
Offline

I would like to see a group experiment, where we start working on just one word. I believe that if we worked hard to get one word sounding convincing, we would make much more progress, than trying to apply so much to an entire song.

Apr 9 2004
03:49
s_j_e
Member
Members
Forum Posts: 5
Member Since:
Feb 28 2004
Offline

Lol – the one word project. Who has the time to focus like that on new software? I just want something I can type in the words, maybe tweak it a bit and generate a decent sounding demo quality voice.

Anything it can’t do with that level of input from me is in the realms of “experimental” and even just as a hobbyist on this stuff, I have other aspects of the music I want to focus most of my time on – like structure and the mix and the instrumental timbres.

Lola is kinda fun but it’s not yet easy enough to use or flexible enough for the big time. My view, anyway.

But hey I’m using it despite all that and appreciate the technology – and that it will hopefully get better over time. Check out my song in the showcase forum.

Steve

Apr 9 2004
09:25
keepspace
Member
Members
Forum Posts: 14
Member Since:
Mar 6 2004
Offline

I agree with the above. We need to get an acceptable result from just entering the words and notes, and then only have to play for hours (perfectionists only) for special variations of expression etc.

Also like above, I’m using Lola anyway because having a singer to sing songs is still a breakthrough for me.

My hopes rest on Miriam and what may follow.

Apr 9 2004
10:42
quetzalcoatl
Member
Members
Forum Posts: 173
Member Since:
Feb 26 2004
Offline

But the point is we’re not getting acceptable results from just entering the words and notes. Vocaloid is not intutive enough for this ease-of-use yet. Don’t waste any more time along this route, stop kidding yourself. Once you’re over the novelty, the fact is we’re having great difficulty getting Vocaloid to pronounce words clearly or realistically. This is because we’re trying to run before we can walk. If we work on entire songs, we’re having to learn too many different “tweaks” on too many words and places in the whole song at once. You may discover that a certain experssion tweak here and a certain phoneme tweak there works for you, but you won’t really appreciate or remember why. If we worked on one-word tutorials that covered for e.g. vowel pronunciation, another one that covered attack, one on getting an American accent and so on, all using just one-word at a time, we would be better equipped to adapt that knowledge to everything else. It would also be quicker, fast rendering and more memorable. The analogy of doing the opposite thing, would be like trying to learn all the design elements of a web page at the same time! You’re in effect trying to learn how to make buttons, how to create a nav bar, how to prepare photos, how to stream video etc. all at the same time. The analogy is all the elements of a complete web page as all the elements of a complete song. All I’m saying is let’s take this one step at a time. If we can get just one word to sound great, it’s a better start than stumbling through making twenty “different” adjustments. Lyrics and musicality are totally different areas, that’s why only a few people do both well. Yet here we are trying to to do both! In Vocaloid, lyrics is a hard enough area, the pronunciations represent a whole universe of learning, even before we get into the musical side of a song. Hell, we are cramming all the learning into one horribly frustrating and disappointing experience.

In most of the demos I’m hearing, I have not heard one instance where I’ve thought, “Oh yes, he’s cracked how to do so and so”. Instead I’m hearing half-arsed attempts at making complete songs. At best they sound like they’re supposed to be amusing us. I’m not amused. I’m not learning anything that I haven’t already learned for myself. Making a complete song with Vocaloid won’t impress even you in the long run. We’re not making progress this way. You’re trying to do the final exam without taking the course.

Wake up and smell the coffee :!:

That’s what I’m talking about 8)

Apr 9 2004
11:29
andromeda
Member
Members
Forum Posts: 188
Member Since:
Feb 27 2004
Offline

I think you get the prize for longest message, Q. Seriously, I can see your point about not walking before you can run etc. I would only criticise the “one word” idea by pointing out that the one word may sound different depending on what comes before and after. It might be better to work on, say, three words. How about “I love you” as I’m sure these have appeared in many a song!
Cheers
Chris

Apr 9 2004
13:34
s_j_e
Member
Members
Forum Posts: 5
Member Since:
Feb 28 2004
Offline

Well I kind agree with you both, Q & A – (hey I like that!). In tweaking anything you should only make one change at a time to really appreciate the impact of the specific change. I broke my vocal passages down into short phrases passages and worked on and recorded each bit (about 4 bars each) separately.

Largely that was driven by the need to get the pre-processing down under a minute for me to be able to put up with it. (v good news that the next release might have shorter processing times)

But the sounds of each syllable are so context-sensitive that I doubt you could really get a meaningful answer about a single word. (mind you, if you have a way of making “touch” sound less like “terch” I’d be glad of some insight). The same phoneme can sound quite different depending on the accent, vibrato, tone, etc as well as what other word or syllable comes before and after. Even if you had a great version (or several great versions) of a word you’d still potentially need to tweak once it was in context.

I’d rather let the Yamaha R&D people invest that kind of time and energy and present me with a better version2. Lol – that’s *their* job.

Steve

Apr 9 2004
14:38
andromeda
Member
Members
Forum Posts: 188
Member Since:
Feb 27 2004
Offline

To get Vocaloid to sing [i:1zamoekk]touch [/i:1zamoekk]the way I think you want, try typing in the phonemes th { tS
It’s more of a [i:1zamoekk]tatch[/i:1zamoekk] but sounds nearer what you expect on [b:1zamoekk]short[/b:1zamoekk] notes. It works for me with Lola on notes just above middle C.
Cheers
Chris

Apr 9 2004
15:13
rickpaul
Member
Members
Forum Posts: 18
Member Since:
Apr 7 2004
Offline

On the one word test idea, I have to agree that probably wouldn’t be helpful, though the notion of a short phrase should be more so. Context really does affect whether something sounds natural or such.

If I were to propose a phrase to try, it might be the familiar, “happy birthday to you,” simply because everyone knows what the melody for that is, whereas something like, “I love you,” as suggested by someone else comes in lots of different melodies, both familiar and otherwise.

That said, the time to actually do that is another consideration altogether. While it is something I definitely wouldn’t mind doing as a learning experience, in all practicality, it will probably be another week or more before I’d actually be able to get around to dedicating the time that would take due to a combination of other real projects and some out of town time off. And, based on all the results I have heard to date, I have to say I’m pessimistic as to the likelihood of real success anyway.

The real question in this, though, is very valid: “Can it be done?” That is, with an infinite amount of work, can Lola (or Leon) be made to sound like a real native-English speaking (be that American or British — I’d be happy with either at this stage) human. Then, if the answer to that question is, “yes,” there is the more important question of whether the knowledge of how to do it with “infinite work” can be applied to doing it with a reasonable amount of work for practical use.

Of course, even as things stand now, there may still be applications for these products in more experimental/electronic types of music that don’t depend on a real human voice. I mean vocoders have been used enough in real projects. But that’s a different application than the “virtual soul vocalist” notion. Right now it’s more like “virtual Eastern European robotic vocalist.” :(

Rick

Apr 9 2004
19:51
quetzalcoatl
Member
Members
Forum Posts: 173
Member Since:
Feb 26 2004
Offline

The point about the context of any single correct words is correct, and learning phrases is also vital, but they are not the best places to start. It’s still running before we can walk. Demo after demo, as soon as I hear more than two words strung closely together I can instantly hear disaster. Vocaloid simply cannot handle it, and it sounds aweful. It’s just not working, and the sooner we’re all honest about it the better.

[quote="s_j_e":25urdko4]I’d rather let the Yamaha R&D people invest that kind of time and energy and present me with a better version2. Lol – that’s *their* job.[/quote:25urdko4]

Well, most of the early adopters have got quite involved with offering tips and ideas on how Vocaloid could be improved. If I could use another analogy .. if Photoshop users had waited for Adobe to show them how to do things, you woudn’t see half of the interesting designs you now see on web sites. The cutting edge stuff has always come from tutorials by users who have discovered interesting new things. The bottom-line is, yes you can wait for Yamaha to improve Vocaloid, or we can get working with what we’re got now. We could be waiting a few years……

Apr 10 2004
17:09
gray
Member
Members
Forum Posts: 304
Member Since:
Feb 27 2004
Offline

I think the problem is more in the pitch range than pronunciation. I can understand most of the words in my files upon lyric input. Its when I add the low brightness, clearness etc. that the pronunciation becomes blurred. Bu the alternative is to have a singer sounding like a computerized Zombie. I feel the biggest problem is the sampling was not done in enough range to cover most songs. I would like to do more Beatles songs, but there just arent many in Leons pitch range. As an example, Led Zeppelin’s What is and what should Never Be that I posted. The lower range vocals sound great. The upper sounds like Leon’s having a bowel movement.

Apr 10 2004
18:59
robotarchie
Member
Members
Forum Posts: 223
Member Since:
Feb 27 2004
Offline

Yeah, ain’t it a bitch. “Context” is a word that crops up again and again. I think the squawker has to be put in the right musical context. The East European robotic thing will probably work. The computerized zombie thing will work. I haven’t heard a soul song context yet that works.

Apr 10 2004
22:21
gray
Member
Members
Forum Posts: 304
Member Since:
Feb 27 2004
Offline

True. Not much soul music in Leons pitch range.

Apr 16 2004
21:52
stoicspartan
New Member
Members
Forum Posts: 1
Member Since:
Mar 5 2004
Offline

Hello…I’m one of the “silent” 143 users mentioned in first post…why am I so silent?…frankly, Lola is the worst 300 dollars I’ve ever spent…I fell for the ads and the hype on the demo front…I should have known better because I’ve been recording since 1971 and was doing midi/computer stuff as soon as it was available…Lola, as many of you point out, is really not much of a vocalist at all…after a few intense days of working with her I realized that she is merely an interesting idea whose time has not arrived…I know that there will be brave users who enjoy the frustrations/challenges presented and will do their best to take this software as far as possible…years from now there may well be a viable product that can actually be of some practical use in (“praxis”as opposed to “theoria”) a working studio …as an early buyer, I registered quite a while back and cruised by today to see what was going on in Lola and Leon Land. apparently things are pretty much as they stood…although the next upgrade at least promises some alleviation of the notoriusly anemic processing speed. Hope no one takes offense here, I’m not bitter nor angry…just thought I’d respond to this becuae it might reveal why so many registered users are not responding…we simply thought the program would deliver in a way that would make it viable in the recording studio, where time and accuracy are imperative. Having realized we wouldn’t find much pragmatic use for this program…we’ve simply moved on. I will continue to remain curious as to the devlopment of this kind of product and will undoubtedly peruse this forum every once in a while. Good luck and clear sailing to those intrepid souls that will be instrumental in pushing this software’s limits and demanding that it live up to its potential!! ……..Respectfully submitted to the registetred users in the Vocaloid community… pro or con…we all have one thing in common…we bought this software!!!!

Apr 17 2004
07:42
andromeda
Member
Members
Forum Posts: 188
Member Since:
Feb 27 2004
Offline

I have to agee with you on this. You echo a number of comments we’ve made on here. Vocaloid is a great idea but it’s not quite ready yet. The brave souls on here who are prepared to spend hours working on a piece are able to get reasonable results. It’s early days for this technology and you are probably right when you say it will be quite a while before it’s possible to get usable results in an acceptable amount of time with not too much effort. A key point is the one about “finding a use for” the product. Even in it’s current form I can find a use for it, but not what I was originally hoping for. I my case I produce mostly instrumental music, so Lola can provide some useful sounds or additions to what I want, so long as I keep my requirements simple and specific. Rather than try to make Lola sing what she can’t sing, the answer, for me, is to write a part for her that is suited to her (limited) capabilities. After all, you don’t write a piano part for playing on a trumpet! The sad thing is that lola is very much that at the moment. The ugly duckling who thinks she’s a swan…..?
Cheers
Chris

This topic is locked No permission to create posts
Forum Timezone: UTC 0

Most Users Ever Online: 108

Currently Online:
2 Guest(s)

Currently Browsing this Page:
1 Guest(s)

Top Posters:

gray: 304

robotarchie: 223

andromeda: 188

quetzalcoatl: 173

Giuseppe: 167

Luka Mitutoyo: 147

Member Stats:

Guest Posters: 1

Members: 7988

Moderators: 0

Admins: 1

Forum Stats:

Groups: 5

Forums: 27

Topics: 764

Posts: 3555

Newest Members: kprem, yang7764, lisajim, gameace, Sunnydoot, Anildoot

Administrators: administrator: 268