Gnuspeech Does Its First Official Release For Free Speech Synthesis

[phoronix]

Phoronix: Gnuspeech Does Its First Official Release For Free Speech Synthesis

The Gnuspeech project has announced its first official release as a speech synthesis from text free software system...

http://www.phoronix.com/scan.php?page=news_item&px=Gnuspeech-First-Release

[uid313]

I would like to hear it.

There is also the free software TTS engine eSpeak which uses formant synthesis, but it is pretty crappy.

[CapsAdmin]

A while ago I was wondering how speech synthesis worked but failed to find anything that was open source or even free so this is cool.

[SystemCrasher]

What the hell? This thing requires proprietary xcode to build?! And only builds on MacOS? Without support of anything else? And THIS thing called GNU? Dear Stallman, WTF? O.O Is it GNU TroLoLo project? Or GNU FirstAprilJoke? Or these days GNU just really not picky on software? GNU/EPIC FAIL.

Edit: looks like I've been looking into wrong tarball with similar name, placed nearby and coming with higher version.

[sharkwouter]

It sounds like Microsoft Sam.

You can easily compile gnuspeechsa on Linux.

[bob l'eponge]

As usual with GNU stuff & team, they have 0 ideas about some word... what it is... wait... mar... mark... marketing, yes, that's it!
Their website comes & looks like the prehistoric world in Lascaux, not a single wav/mp3 file to listen for their new wonderful system, while the complete industry has moved to unit selection voices (www.neospeech.com for an example, or www.cereproc.com), they still promote a technic that's 20 years old. At least, if you want opensource speech synthesis, you've MaryTTS that works, and it's not based on mammouth technology.

[sharkwouter]

For the people wondering, if you can download, compile and test this on Linux with the following set of commands:

wget https://ftp.gnu.org/gnu/gnuspeech/gnuspeechsa-0.1.5.tar.gz
tar -xvf gnuspeechsa-0.1.5.tar.gz
cd gnuspeechsa-0.1.5/
mkdir build
cd build
cmake ..
make
touch trm_param_file.txt
./gnuspeech_sa -c ../data/en/ -p trm_param_file.txt -o output_file.wav "your text goes here"

You can find the result in gnuspeechsa-0.1.5/build. You can of course put in your own text. For some reason you need to manually create a file to be used as trm param file. You can later use this trm param file to recreate the same wav file with the gnuspeech_sa_trm binary.

You can change which voice is used in gnuspeechsa-0.1.5/data/en/trm_control_model.config

[schmidtbag]

I would like to hear it.

There is also the free software TTS engine eSpeak which uses formant synthesis, but it is pretty crappy.

I've tried espeak before. It's not that bad, you just have to try different voices and accents. I found the female voices are much easier to understand.

[SystemCrasher]

you've MaryTTS that works, and it's not based on mammouth technology.

It seems to be based on Java. That what makes it nearly useless for most practical purposes.

[TheCycoONE]

I've tried espeak before. It's not that bad, you just have to try different voices and accents. I found the female voices are much easier to understand.

At this point in time a TTS engine shouldn't be remotely hard to understand. The good ones are not only easy to understand but can project emotion - they also cost a small fortune which isn't helped by Nuance's purchase of Loquendo.

For personal use, yeah you can make your computer tell you stuff with eSpeak, and you can probably tell what it's saying, but for professional use it's a non-starter - even 5 years ago. As a developer at a small company that relies on TTS technology and is still using AT&T natural voices due to prohibitive licensing costs for modern voices - it would be great if there was a competitive FOSS alternative. eSpeak is not it; maybe this is, but Bob makes me doubtful.