Festival: Text-To-Speech for Puppy Tahr

[ariel]

This is Festival V. 2.1 for Puppy Tahr + English voices. A Text To Speech (TTS) program converts normal language text into speech with a synthesized voice.

In order to see the voices included in the package type in a terminal:

Code: Select all

root#festival

festival>(voice.list)

and to change the use of a voice in place of the standard one type:

Code: Select all

festival>(voice_VOICE_NAME)

For the download visit https://lab2000-linux.homepc.it and go to the download section.

********** ************** ****************

Here are some samples of the available standard English voices:


don_diphone
ked_diphone
rab_diphone
kal_diphone
us1_mbrola
us2_mbrola
us3_mbrola
en1_mbrola
cmu_us_slt_arctic_hts

SAMPLES OF THE NITECH HTS VOICES
========================

nitech_us_awb_arctic_hts
nitech_us_bdl_arctic_hts
nitech_us_clb_arctic_hts
nitech_us_jmk_arctic_hts
nitech_us_rms_arctic_hts
nitech_us_slt_arctic_hts

********** ************** ****************




here is the link for the download of the Tcl/Tk script for reading and converting .txt files into speech or mp3 files.

[musher0]

Nice site, ariel!

[Flash]

What is TTS?

[ariel]

thank you musher0

I hope that the website may be useful for the community. For this reason if somebody wants to share a program but doesn't have a place to keep their files please PM me and I will upload the program on the site !


@Flash

You're right, perhaps not everybody knows what TTS is. I've added a small explanation in the first post.

[nic007]

Hi, ariel. Can you post a small audio file as illustration how these voices sound? I'm using 2nd Speech Center with the American TruVoice running in Wine. Attached is a sound sample of the result I get.

[ariel]

hi nic007,

I have uploaded some samples for the available English voices. Please see the first post.The text read in the samples is the following:

"We've told yarns by the camp-fire in the prairies; and dressed one
another's wounds after trying a landing at the Marquesas; and drunk
healths on the shore of Titicaca. There are more yarns to be told, and
other wounds to be healed, and another health to be drunk. Won't you let
this be at my camp-fire to-morrow night? I have no hesitation in asking
you, as I know a certain lady is engaged to a certain dinner-party, and
that you are free. There will only be one other, our old pal at the
Korea, Jack Seward. He's coming, too, and we both want to mingle our
weeps over the wine-cup, and to drink a health with all our hearts to
the happiest man in all the wide world, who has won the noblest heart
that God has made and the best worth winning. We promise you a hearty
welcome, and a loving greeting, and a health as true as your own right
hand. We shall both swear to leave you at home if you drink too deep to
a certain pair of eyes. Come!

I hope that you can find these tools useful and that you can switch to the linux voices

Please let me know what you think of them.

[Pete]

I'm sorry to say, but those TTS voices sound really bad.
Why not use Google text to speech?
Much more natural sounding.

It's available as a browser plug-in and for Android.
If I'm not mistaken, it will work "out the box" with the latest Chrome as well.

As an example, I took nic007's last post and ran it through text to speech, selected UK female and the result is attached below.

Note fake zip extension.

[LazY Puppy]

I'm sorry to say, but those TTS voices sound really bad.

Yes, they really sound very very bad (to not to say: ugly).

Also they seem to have some digital 'clippings'.

Is it possible to create voices/samples from my own recorded voice to use that instead of those horrible voice-sounds?

[Pete]

@LazY Puppy

Although in theory, it seems like all that one would need would be a text parser which in turn uses a look-up table of words (e.g. a dictionary file of pre-recorded words), in practice it is very different if it is to sound realistic and pleasant.

For example, if a word is followed by a question mark, people normally will pronounce that word differently to say if it was in the middle of a sentence or followed by an exclamation mark.

This is further complicated by languages such as German and Hungarian that use umlauts and Latin based languages that use accents, all changing the way a word is pronounced.

Back in the early 80's there were systems (for games) that used pre-recorded generic words, but as you can imagine, it was limited and had no "expression", sounding very monotone and unrealistic.

Then there are numbers, for example the text "2016", it would have to "spoken" as two thousand and sixteen.
Can you imagine how many combinations there are even if you record the thousands, hundreds, tens and units only.

Then there are equations, for example, 10 X 2 = 20
The software needs to know that X in this case is "multiplied by" and not just simply a capital "X".
How about this?
x = 20 X 2y

The software would also have to deal with things like:
1st, 2nd, 3rd, 4th,........

Speech synthesis is a very complicated thing and the subject of much research, have a look at this page:

https://en.wikipedia.org/wiki/Speech_synthesis

[nic007]

Those festival voices sound horrible. Lots of work to be done there for improvement.

[ariel]

Ok, for those who are fond of horror, here is a script that makes use of the English voices packeged with Festival and can read/save as mp3 a text in the clipboard or a .txt file by using a GUI.

The script uses Tcl/Tk scripting language for portability.


for the download see first post.

[ariel]

New voice for Festival and Puppy Tahr packaged.

It's a CMU Arctic voice. I think that this is an example of the best available voices on the market that one can have for free. But I like horror so have a look for yourself at the sample in the first post.

here is the download link.

[ariel]

Nitech HTS voices patched and packaged for use with Festival 2.1. Hear the samples of these voices linked in the first post


Please note that these voices have a better quality than the standard ones.

Download at https://lab2000-linux.homepc.it in the download section of Puppy Tahr.