This project and the documentation from my blog is now hosted on GitHub the links in the post have been changed accordingly.
What is Maryspeak
I’ve been inspired by Ken Fallon via Hacker Public Radio to write maryspeak; a small java program to make the core features of Mary Text To Speech readily accessible as a shell command. The aim of maryspeak is to reduce the friction for Linux shell users to use MaryTTS. It accepts text input via the command line, via a file or via stdin, processes it using MaryTTS, then outputs speech via sound, to a file, or to stdout. It also allows for the selection of a voice and/or a MaryTTS server.
MaryTTS is written in Java and the UI is not transparently accessible to the Linux command line. maryspeak is a wrapper around the MaryInterface classes used by the MaryTTS Java and http clients. Because its written in Java there is no reason this cannot also be on a Windows system, but the command switches are closer to the GNU conventions.
Installing Maryspeak
This installation assumes Debian but the same principles will apply to other distros.
Using Maryspeak
You can use maryspeak standalone or against a MaryTTS server.
- Show the full Maryspeak usage instructions with -h or –help
$ maryspeak --help
Further Exploration/Exploitation of MaryTTS
Maryspeak only offers a subset of the functionality of MaryTTS. Maryspeak is also not using the streaming capabilities of the MaryTTS library, so processes things in series: gather input, process the input, then output audio.
MaryTTS has a rich depth of functionality beyond that used by Maryspeak. For serious use I would recommend investigating this.
MaryTTS can be used directly from command line if required via:
$ java -cp "classpath" [properties] marytts.client.http.MaryHttpClient [inputfile]
Assuming MaryTTS is installed as per my previous post, the instructions for usage of this class can be obtained by compiling and running the following java class:
public class ShowUsage {
public static void main(String[] args) {
marytts.client.http.MaryHttpClient.usage();
}
}
For your convenience I have done this, for MaryTTS version 5.1, this is the output:
usage:
java [properties] marytts.client.http.MaryHttpClient [inputfile]
Properties are: -Dinput.type=INPUTTYPE
-Doutput.type=OUTPUTTYPE
-Dlocale=LOCALE
-Daudio.type=AUDIOTYPE
-Dvoice.default=male|female|de1|de2|de3|...
-Dserver.host=HOSTNAME
-Dserver.port=PORTNUMBER
where INPUTTYPE is one of TEXT, RAWMARYXML, TOKENS, WORDS, POS,
PHONEMES, INTONATION, ALLOPHONES, ACOUSTPARAMS or MBROLA,
OUTPUTTYPE is one of TOKENS, WORDS, POS, PHONEMES,
INTONATION, ALLOPHONES, ACOUSTPARAMS, MBROLA, or AUDIO,
LOCALE is the language and/or the country (e.g., de, en_US);
and AUDIOTYPE is one of AIFF, AU, WAVE, MP3, and Vorbis.
The default values for input.type and output.type are TEXT and AUDIO,
respectively; default locale is en_US; the default audio.type is WAVE.
inputfile must be of type input.type.
If no inputfile is given, the program will read from standard input.
The output is written to standard output, so redirect or pipe as appropriate.
So for a quick demo try putting some text into a file test.txt (use a full stop at the end of your text or Mary speaks slowly for some reason) then run:
$ java -cp "/opt/marytts-5.1/lib/*" marytts.client.http.MaryHttpClient test.txt | aplay
Note: that the output is piped into aplay which can play back a .wav stream. If you don’t have aplay installed you can > output to a file.wav to play later.
Much more can be done using the full MaryTTS libraries within a Java program.
The Linux Foundation ‘Introduction to Linux’ course is now available via the edX online training system.
