HOWTO: Make festival TTS use better voices (MBROLA / CMU / HTS)

[Tycho451]

Hi,
first: Thanks for this great Howto!

I have a problem though.
When I try to use the enhanced Nitech HTS voices I get kicked out.

$ festival
Festival Speech Synthesis System 1.96:beta July 2004
Copyright (C) University of Edinburgh, 1996-2004. All rights reserved.
For details type `(festival_warranty)'
festival> (voice_nitech_us_slt_arctic_hts)
nitech_us_slt_arctic_hts
festival> (SayText "Incoming message")
Aborted
$

I tried to compile from source (although I already had 1.96:beta), but the speech-tools don't compile.

g++ -c -g -O3 -fPIC -Wall -Wno-non-template-friend -I../include -DINSTANTIATE_TEMPLATES est_file.cc
../base_class/EST_TNamedEnum.cc: In member function »EST_read_status EST_TNamedEnum<ENUM>::priv_load(EST_String, EST_TNamedEnum<ENUM>*) [with ENUM = EST_EstFileType]«:
est_file.cc:72:   instantiated from here
../base_class/EST_TNamedEnum.cc:215: Fehler: keine passende Funktion für Aufruf von »split(EST_String&, EST_String [12], int, EST_Regex&, char)«
../base_class/EST_TNamedEnum.cc:239: Fehler: keine passende Funktion für Aufruf von »split(EST_String&, EST_String [12], int, EST_Regex&, char&)«
make[2]: *** [est_file.o] Fehler 1
make[1]: *** [utils] Fehler 2
make[1]: Verlasse Verzeichnis '/home/marco/fest_tmp/speech-tools-1.2.96~beta'
make: *** [build-stamp] Fehler 2
debuild: fatal error at line 1311:
couldn't exec fakeroot debian/rules:

So I just compiled and installed festival from source. Didn't make a diffrence though.

Has anyone any ideas what might be wrong?

[shane2peru]

Wow, great how too!!! Excellent!

Shane

[Tycho451]

I also tried to get german festival voices with this guide:
http://docs.kde.org/stable/en/kdeacc...erman-festival

I do everything according to the guide, but I can't get it to work (SIOD ERROR: unbound variable : voice_german_de2_os).

[shane2peru]

ok, I'm having real problems with the mbrola side of things. I have used espeak in the past, and it works well. I have used it with mbrola as well. However now whatever I do, mbrola is not working correctly. I'm working on Ubuntu Intrepid 64 version, all up to date. MBROLA is not cooperating. I downloaded and installed the mbrola binary for 64 bit from here: http://tcts.fpms.ac.be/synthesis/mbrola.html - the last one in the list. I unzip it, I chomd +x the mbrola file (to make it executable) then copy it to /usr/local/bin. Then I run the mbrola to make sure it works, and sure enough it gives me the speal, it is installed and running. When I then run the test according to this page: http://tcts.fpms.ac.be/synthesis/mbrola/mbruse.html and it doesn't work. Any help that can be provided would greatly be appreciated. I do have the voices downloaded, and when I run the test line with the appropriate directories, it doesn't seem to work for me. It is driving me crazy. Any help would greatly be appreciated.

Shane

PS festival is working, however I want the Spanish voices that come with mbrola, and I didn't see them in any of the others. I also would just like to have mbrola working correctly.

[MonkeeSage]

Hi,
first: Thanks for this great Howto!

I have a problem though.
When I try to use the enhanced Nitech HTS voices I get kicked out.

Do any other voices work properly (other Nitech voices, Mbrola voices, &c) or do they all die like that? Did you set up any special output method in the config file?

[MonkeeSage]

I also tried to get german festival voices with this guide:
http://docs.kde.org/stable/en/kdeacc...erman-festival

I do everything according to the guide, but I can't get it to work (SIOD ERROR: unbound variable : voice_german_de2_os).

It looks like to use the ims_german_festival requires installing festival / speech tools from vanilla source. I'm not sure how easy it would be to get it working with this guide. Might be pretty complicated. I would suggest following their README and installing that copy of festival in /usr/local or /opt. Then you can have a German festival install and the English voices from this guide. Good luck!

[MonkeeSage]

ok, I'm having real problems with the mbrola side of things. I have used espeak in the past, and it works well. I have used it with mbrola as well. However now whatever I do, mbrola is not working correctly. I'm working on Ubuntu Intrepid 64 version, all up to date. MBROLA is not cooperating. I downloaded and installed the mbrola binary for 64 bit from here: http://tcts.fpms.ac.be/synthesis/mbrola.html - the last one in the list. I unzip it, I chomd +x the mbrola file (to make it executable) then copy it to /usr/local/bin. Then I run the mbrola to make sure it works, and sure enough it gives me the speal, it is installed and running. When I then run the test according to this page: http://tcts.fpms.ac.be/synthesis/mbrola/mbruse.html and it doesn't work. Any help that can be provided would greatly be appreciated. I do have the voices downloaded, and when I run the test line with the appropriate directories, it doesn't seem to work for me. It is driving me crazy. Any help would greatly be appreciated.

Shane

PS festival is working, however I want the Spanish voices that come with mbrola, and I didn't see them in any of the others. I also would just like to have mbrola working correctly.

Hi Shane,

Works for me...

wget http://tcts.fpms.ac.be/synthesis/mbrola/dba/es1/es1-980610.zip
unzip -x es1-980610.zip
mbrola es1/es1 es1/TEST/example1.pho test.wav
play test.wav
rm test.wav

What problem are you having with it, exactly?

Jordan

[MonkeeSage]

Can you suggest me anything what else to do to make this voice usable.
thanks in advance, sasa.

Hi Sasa,

Alright, I finally got some time to play with this. Here's what I found. I got the voice to load by editing /usr/share/festival/voices/croatian/cmu_cr_tp_clunits/festvox/cmu_cr_tp_clunits.scm and changing lines 40-41 from:

(if (null (member 'clunits provided))
    (load (path-append libdir "../src/modules/clunits/acost.scm")))

to:

;;(if (null (member 'clunits provided))
;;    (load (path-append libdir "../src/modules/clunits/acost.scm")))

But now when I try to synthesize some text, I get e new error:

festival>(voice_cmu_cr_tp_clunits)         
cmu_cr_tp_clunits
festival> (SayText "Lijepa nasa domovino") 
SIOD ERROR: Undefined SynthType Cluster

I have no idea how to fix this.

I only tried it with 1.96, and this voice is really old (made in 2000 according to the festvox server), so maybe it would work with 1.4. Sorry I can't be more helpful!

[shane2peru]

Hi Shane,

Works for me...

wget http://tcts.fpms.ac.be/synthesis/mbrola/dba/es1/es1-980610.zip
unzip -x es1-980610.zip
mbrola es1/es1 es1/TEST/example1.pho test.wav
play test.wav
rm test.wav

What problem are you having with it, exactly?

Jordan

Ok, I ran all of these, and I don't get any errors. Except instead of using play I use vlc to play the wav file. I am using oss instead of alsa, and play doesn't work with oss. I have tested vlc on other wav files and it works fine. However when I use vlc test.wav all I get is this horrible sound that someone is running a fan into the microphone. Very loud and bad. It doesn't give me any error reports.

Shane

[MonkeeSage]

Ok, I ran all of these, and I don't get any errors. Except instead of using play I use vlc to play the wav file. I am using oss instead of alsa, and play doesn't work with oss. I have tested vlc on other wav files and it works fine. However when I use vlc test.wav all I get is this horrible sound that someone is running a fan into the microphone. Very loud and bad. It doesn't give me any error reports.

Shane

Well, the only difference seems to be our mbrola binaries (I have a 32-bit CPU). Can you run the 32-bit mbrola binary?