HowTo Text to speech for the kindle keyboard

[knc1]

Originally Posted by twobob

Will certainly be looking into what is required to make a "better" voice.

1) An audio lab (ah, you have one, sorry forgot)
2) One or more voice models (ah, sorry again, you have access to the entertainment industry over there)
3) Time

Anybody want to donate some time? He needs another ten hours a day - with a corresponding supply of tea.

[twobob]

Originally Posted by knc1

1) An audio lab (ah, you have one, sorry forgot)
2) One or more voice models (ah, sorry again, you have access to the entertainment industry over there)
3) Time

Anybody want to donate some time? He needs another ten hours a day - with a corresponding supply of tea.

Even just the free tea would fine

[twobob]

Tiny little oddnesses... -voice appears to be playing at half time in most cases at the moment. on the k3.
IIRC this was something to do with the card itself.
(GM and I were going to exploit this odd reaction to 16k mono as it uses almost no processor for as-yet no fully-known reasons )

for example on a k3.
[root@kindle root]# flite -voice rms -t "Hello... Hello... This is the voice of the mysterons we know you can hear us earthlings"

will sound how you might imagine that would... slow and low.

Let's render it..

[root@kindle root]# flite -voice rms -t "Hello... Hello... This is the voice of the mysterons we know you can hear us earthlings" /mnt/us/music/mysterons.wav

and play it again to check the formats etc.. (and that it still sounds funky)
[root@kindle root]# aplay /mnt/us/music/mysterons.wav

Playing WAVE '/mnt/us/music/mysterons.wav' : Signed 16 bit Little Endian, Rate 16000 Hz, Mono

Yep... low and weird.

Let's shove it on the Kindle Touch (via the pc)

me@PC ~ $ scp root@k:/mnt/us/music/mysterons.wav ./; scp ./mysterons.wav root@kt:/mnt/us/music/mysterons.wav

mysterons.wav 100% 163KB 162.7KB/s 00:00
mysterons.wav 100% 163KB 162.7KB/s 00:00

and play it again...

[root@kindle root]# aplay /mnt/us/music/mysterons.wav

Playing WAVE '/mnt/us/music/mysterons.wav' : Signed 16 bit Little Endian, Rate 16000 Hz, Mono

now it sounds completely normal.

So... Conclusion? 16k mono files freak out on the k3 OOTB



One more point... inside one the voices awb_time there is a resource reference missing/malpointed.

[root@kindle root]# flite -voice awb_time -t "Do I work?"

clunits: unit type "pau_ax" not found
clunits: can't find tree for pau_ax
clunits: unit type "pau_ax" not found
etc...

So due diligence done. issues reported.

The file attached below should sound slow on a 3 and normal on a 5. Thanks.

[knc1]

Sampling rate on the K3 off by one (binary magnitude) ?

I don't know audio files well enough to say if you could edit the metadata in one to compensate for the incorrect sample rate calculation.
But if you can, that might be one way to find what is happening to that file on the K3.
From that information, maybe a work-around ? ? ?

[twobob]

[root@kindle root]# flite -voice kal -t "This is the voice of the misterons we know you can hear us earthlings" /mnt/us/test1.wav;\
aplay /mnt/us/test1.wav

Playing WAVE '/mnt/us/test1.wav' : Signed 16 bit Little Endian, Rate 8000 Hz, Mono

works as expected. even on the 3... as it (the voice kal) is 8k by default.

Only 16k does this oddness

workaround? figure out how to force to other voices to render at a different rate perhaps

EDIT: since I failed to do this. perhaps it's a compile time option. will check at some point. in the interim streaming at a different rate would probably do the job. will investigate some options

[Lupines]

Originally Posted by twobob

[root@kindle root]# flite -voice kal -t "This is the voice of the misterons we know you can hear us earthlings" /mnt/us/test1.wav;\
aplay /mnt/us/test1.wav



works as expected. even on the 3... as it (the voice kal) is 8k by default.

Only 16k does this oddness

workaround? figure out how to force to other voices to render at a different rate perhaps

EDIT: since I failed to do this. perhaps it's a compile time option. will check at some point. in the interim streaming at a different rate would probably do the job. will investigate some options

I'll upload the modified source files in case someone needs them over the weekend. I will also try to work on trying to make the binary smaller. 20MB is way too big. I believe the libs for alsa can be removed from the binary to make it smaller. I've been really busy and with no free time. After having a decent binary the other thing would be to add more voices. I haven't looked into this, yet.

Lupines.

[twobob]

Originally Posted by Lupines

I'll upload the modified source files in case someone needs them over the weekend. I will also try to work on trying to make the binary smaller. 20MB is way too big. I believe the libs for alsa can be removed from the binary to make it smaller. I've been really busy and with no free time. After having a decent binary the other thing would be to add more voices. I haven't looked into this, yet.

Lupines.

sounds good.
will look out for it. much obliged!

[Lupines]

Originally Posted by twobob

sounds good.
will look out for it. much obliged!

I tried a compile with -O3 and another with -Os. Also removing the CFLAGS=-DEMBEDDED from configure. I get the same results. You are more than welcome to give the compiling a try and play with it. I just uploaded the source files here:

flite-1.4-release-arm-src.tar.gz

http://ge.tt/86zHm1X?c

Lupines

[twobob]

grabbed and appreciated.

[twobob]

Did something with it.

http://www.mobileread.com/forums/sho...50&postcount=7

Thanks. will play more.

[Lupines]

Originally Posted by twobob

Did something with it.

http://www.mobileread.com/forums/sho...50&postcount=7

Thanks. will play more.

HA! Nice... At least write down my nickname correctly! hehehe

Also I wouldn't mind if you mention the guys: Alan W Black and Kevin A. Lenzo
Speech Group at Carnegie Mellon University.

I know I put all the files that came with the source files in the .tar.gz, I just don't want to get credit for something I didn't code myself. Mine was just a repacking effort, that's all.

Lupines.

[twobob]

Originally Posted by Lupines

HA! Nice... At least write down my nickname correctly! hehehe

Also I wouldn't mind if you mention the guys: Alan W Black and Kevin A. Lenzo
Speech Group at Carnegie Mellon University.

I know I put all the files that came with the source files in the .tar.gz, I just don't want to get credit for something I didn't code myself. Mine was just a repacking effort, that's all.

Lupines.

Done. Credit spread around. Name unmangled.
http://www.mobileread.com/forums/sho...50&postcount=7

[Lupines]

I started working on festival.

http://www.festvox.org/packed/festival/2.1/

I am having lots of problems while compiling. But I was able to get a binary out of it. The problem I am having right now is after using the festival binary, I get:

festival> (SayText "Hello world")
SIOD ERROR: unbound variable : SayText

Lupines.

[twobob]

Originally Posted by Lupines

I started working on festival.

http://www.festvox.org/packed/festival/2.1/

I am having lots of problems while compiling. But I was able to get a binary out of it. The problem I am having right now is after using the festival binary, I get:

festival> (SayText "Hello world")
SIOD ERROR: unbound variable : SayText

Lupines.

http://forums.asterisk.org/viewtopic.php?f=1&t=72856

The main problem was I downloaded the latest version of Festival in /usr/local/src/Festival and built it.
And there is a festival.scm in the /usr/local/src/festival/lib and I was adding the above code in that file.
But Festival for some reason always picks up the scm file from /usr/share so you have to make the scm chnages there even if you downloaded a new Festival version.

Dont know why the Asterisk folks don't keep this code in the by default festival.scm in the /usr/share location. It would have been just plug and play.
Can you someone explain the reason just for my FYI.

[Lupines]

Originally Posted by twobob

http://forums.asterisk.org/viewtopic.php?f=1&t=72856

I tried that. It isn't working somehow. I put all the .scm files on /usr/share/festival, /mnt/us/arm-linux/share/festival, and also under /mnt/us/arm-linux/share/festival

I will check if I am doing something wrong at compiling time. If you want to have a look, let me know and I will upload what I've done so far. I don't mind credit, I want stuff that works. If you are interested in this, I can guide you on the modifications I've done to this. Or you can just run diff against the original source code.

Lupines.

P.S.: Is there an irc server/channel people around here hang out?