2017-09-06 04:34:44

This is quite new.
http://lyrebird.ai

2017-09-06 13:08:11

Fun stuff. I remember people talking about it a couple of months ago.

To see a world in a grain of sand, and a heaven in a wild flower.
Hold infinity in the palm of your hand, and eternity in an hour.
William Blake - Auguries of Innocence, line 1 to 4

2017-09-06 15:07:01

Will this make a SAPI voice? If so, I really am going to check this thing out!

The Beast continued its studies with renewed Focus, building great Reference works and contemplating new Realities. The Beast brought forth its followers and acolytes to create a renewed smaller form of itself and, through Mischievous means, sent it out across the world.
from The Book of Mozilla, 6:27

2017-09-06 16:22:19

As far as I know, right now the only thing you can do with the voices you make is to have their site generate sentences you enter, then play them back, and share those links on social media. You can't download the generated files or voices. Also, their evaluation agreement specifically forbids the generation of voices of third parties I.E. people other than yourself. Besides, you'd have to get them to read specific sentences they specify.

2017-09-06 20:04:36

is there a way I can make this stop throwing upload errors at me? I've had to re-record a sentense about five times now and its still not getting through. I'm telling you, its not my net, because its fine.

I used to be a knee like you, then I took an adventurer in the arrow.

2017-09-06 20:19:32

Other than just being a gimmick, it doesn't sound all that useful to me.

Now if it could take my voice and create a TTS voice that I could use in JAWS or SAPI. That I'd consider useful and worth paying for.

2017-09-06 20:34:13

wow, it sounds like my voice, funny.

2017-09-06 23:31:33

I'm interested in seeing if everyone's voice sounds like Microsoft David in inflection like mine does. It seems like no matter how you sound, it makes your inflection basically the same way.
https://lyrebird.ai/g/fDGBtvIl

Discord: dangero#0750
Steam: dangero2000
TWITCH
YOUTUBE and YOUTUBE DISCORD SERVER

2017-09-06 23:48:54

well yours ended up far better than mine did, lol probably due to me using a kinda crappy microphone, on top of not being very good at spontaneous sentence reading so I had a few pauses in there

"You know nothing of death... allow me to teach you!" Dreadlich Tamsin
Download the latest version of my Bokura no Daibouken 3 guide here.

2017-09-07 00:24:00

Can I hear yours, I'm interested in hearing other people's recordings.

Discord: dangero#0750
Steam: dangero2000
TWITCH
YOUTUBE and YOUTUBE DISCORD SERVER

2017-09-07 00:37:49

nope, nobody's hearing mine... it's awful, like actually

"You know nothing of death... allow me to teach you!" Dreadlich Tamsin
Download the latest version of my Bokura no Daibouken 3 guide here.

2017-09-07 00:59:12

Everyone will have a bad voice at 30 sentences or so. For example, i tried to right, "I am a robot from the planit known as Sattern. I can withstand temperatures up to 48000000000 degrees, and if I get down to -20 my systems will go weeeeeeeeeeeeeeeeeeeeeeeeeee!" Instead, it came out like this. "I am a robot from the planit known as Sattern. I can withstand temperatures up to 4 8 0 0 0 0 0 0 0 0 0 degrees, and if I get down to -20 my systems will go to we!" Try recording more sentences and see how it turns out afterwords.

Discord: dangero#0750
Steam: dangero2000
TWITCH
YOUTUBE and YOUTUBE DISCORD SERVER

2017-09-07 08:44:13 (edited by AlexN94 2017-09-07 08:45:05)

Because I decided to record them, just so I have them locally too... This is, as the file name says, after 30 sentences... Have some Robert Frost:
https://dl.dropbox.com/s/etd9xsulei6gji … rdings.wav

I'm up to 90 recordings so far, not much of a difference, at least not for that bit of text.

To see a world in a grain of sand, and a heaven in a wild flower.
Hold infinity in the palm of your hand, and eternity in an hour.
William Blake - Auguries of Innocence, line 1 to 4

2017-09-07 09:30:10

I swear I broke my account. Tried to create my voice, didn't work. When I try to type in text in generate a sample, it just says something went wrong, please try again. Tried to delete my voice, and recreate it. Tried using three web browsers on two computers, tried logging out and back in. Tried deleting both my voice and my recordings and making 30 new recordings and creating the voice from those. Still no dice. I really have no clue what I did, but this is pretty darn discouraging after having heard at least a dozen other people show off their voices. I put a lot of effort into making sure the diction was the way I wanted it, and the recordings sounded the way I wanted in terms of sound quality, microphone placement etc. Meh. I'll keep trying and see if I can get this to work.

Make more of less, that way you won't make less of more!
If you like what you're reading, please give a thumbs-up.

2017-09-07 17:38:31

I don't know, why It happened.
Maybe You must do something to set Your microphone.
Also, here is my sample of My voice. Don't be desperated, sometime your problem will solved. Just try to contact developers.
And, here is My sample.
https://lyrebird.ai/g/vGBxkKhY

Ja volim samo kafu sa Rakijom.

2017-09-07 19:05:59

Why can't I record? I'm using Safari on macOS 10.12.6. When I press VO space or the space bar on the start recording button and speak the sentence, nothing happens. I pressed the same command again to stop recording. However, the text on the webpage doesn't change. I've even tried routing the mouse to the VoiceOver cursor and pressing VO shift space. What am I doing wrong?

Grab my Adventure at C: stages Right here.

2017-09-07 23:34:37

Try using chrome. I know safari doesn't play nice with scripted sites of this kind. For one, it never worked with facebook audio calls.

2017-09-08 00:10:49

Here is mine This was using a crappy internal mic that would cut offf certain sounds, and my bad reading off some of the longer sentences.

Facts with Tom MacDonald, Adam Calhoun, and Dax
End racism
End division
Become united

2017-09-08 04:12:58

Chrome worked like a charm. I've only recorded about 31 or 32 phrases. Does the voice still sound this robotic even if you record up to a maximum of 300 phrases? Also, this voice may not be the best quality due to my stutter on a few sounds like "M".
https://lyrebird.ai/g/bHcGNQyi

Grab my Adventure at C: stages Right here.

2017-09-08 04:51:04

The better you are at spontaneous sentence reading obviously the better it'll sound, but perhaps the inflection can't make any huge changes from here. Try it though and see what happens.

2017-09-08 10:54:59

I contacted developers about my issue. Seems the problem was that I had reached an API limit. Apparently when I first tried to create my voice, the servers were getting pounded. My repeated attempts to get things to work were going through but kept giving errors before I could get results. Not being aware of this I kept trying until I had reached the limit and the site never told me as such. Thankfully, the limit is reset every 24 hours, and those people who told me it used to give them a lot of errors and hang-ups have said it's running more smoothly now. The limit was graciously increased too. Since then, I've had no problems.

I'm on 180 sentences so far. What's interesting is that I tried so hard to get the diction and intonation the way I wanted it for every recording that it got a fair amount from the first 30 sentences, and adding more recordings only seemed to make small refinements. With some users, this may not be the case; they recommend you do at least 100 for the best-sounding voice. I was expecting a more dramatic improvement, but even after 180 sentences, my voice hasn't changed too markedly. The diction is a little more fluid though and it's starting to catch onto some of the instinctive transitions I do between words in some groupings.

Because some people's voices tend to sound a little broken due to pauses, background noise, room echo and the like, I made an effort to minimize all of that. I tried only to pause at punctuation when recording. And I can fortunately make my bedroom pretty quiet when I have to. I put an fx chain on my mic to pretty much kill all the background noise and some multiband compression to give it a bit more bass and treble. I got insanely picky about my recordings, probably way too picky for my own good since a lot of my efforts would be lost in the coding and resynthesis, but ah well. Couldn't take any chances! smile

Here is a sample of 30 sentences, and another one at 150 sentences. What I've been doing is stopping at every 30 sentences to check the progress. This serves two purposes: 30 sentences is a not-too-small, but not-too-big load to record in one sitting for trying to memorize and read some of those hard sentences, and it lets me catalog the voice's progression. Once I've reached the maximum of 300 sentences, I will post a sample which *attempts* to show the progression of my voice at every stop along the way.

A friend of mine from South Africa also wants to make a Lyrebird voice. Him and I are just as picky about speech sounds and dialects and the like. Because Lyrebird's FAQ sstates that the synthesis is meant for American English, we're both curious as to how it will treat his accent. I've heard how Lyrebird can capture and mimick the phonemes of an accent, for example if a British person records a voice, their synth voice will still have British qualities. This is probably because the engine is analyzing the recordings and storing the shapes of individual phonemes, which it can string together when synthesizing speech. But accents are more than just simple A to B transformations. Many accents, including South African English, have different rules than American English about how to string certain sounds together. The A in enhance is spoken markedly differently from the A in rant, for example. In the former case with enhance, the A is a much open sound, akin to an ah but not quite. In the latter case with rant, it is the opposite and a lot more closed sounding, so it sounds somewhat like rent. Another interesting case will be with words like roll and rolling. In roll, the oal sound is used, at least he uses it, but when another syllable is added to make rolling, the syllable split is different than in American English. WE split between the l and the i. IN other words the first syllable is roll, the second is ing. In South African English and other accents like it, they split between the o and the l, so the first syllable in his case would be row, and the second is ling. The oal sound never happens in this case.

One more interesting case will be with words like jar. The R sound never happens, at least not a rhotic one, but it does happen in the middle of a word like jarring. I can see Lyrebird messing these rules up and more, at least initially, and we will both be greatly amused to hear what it makes of those words. Maybe that is an excellent argument for recording as many sentences as possible. It can catch patterns like that. But then again maybe it won't, since there has to be a line between its pronunciation rules and yours.

I do know in some cases my voice pronounces words in ways that I never did in recordings, for example details. Most people stress the second syllable, I often stress the first particularly when talking about details as nouns. So I might say that I found some interesting deetails. Yeah, I'm strange. I said deetails on at least one recording but it still says details with the stress on the second syllable, which I think sounds a little strange on my voice, but other people probably think my synth voice sounds more normal than I do in person. Lol

Make more of less, that way you won't make less of more!
If you like what you're reading, please give a thumbs-up.

2017-09-08 11:12:59

Your voice in Lyrebird is better, like my voice.

Ja volim samo kafu sa Rakijom.

2017-09-08 15:00:33

I want to delete my samples and start again when my G930's get here, I'm told they don't have the best mic, but anything is better than this laptop mic that tends to do weird noise cancellation and cuts hard consonants. The odd thing is that if I turn off the enhancements on it, it has weird phasing issues where there is literally no center channel, and if you made a monaural recording with it, you literally wouldn't hear anything, only in stereo, and it sounds weird.

I was able to fix this with some bother one time before I reinstalled the system, the actual software for the internal sound card is not accessible, though somehow I managed to fumble around with it and get it into a voice mode or something. I can't exactly remember, as OCR on that screen didn't really yield too much.

Facts with Tom MacDonald, Adam Calhoun, and Dax
End racism
End division
Become united

2017-09-08 17:38:14

@raygrote wow, I like that 130 sentence sample!

2017-09-08 18:15:19

I will do that myself when I will switch on my new laptop!

I am myself and noone is ever gonna change me, I am the trolling master!