GitPlanet

Cheap and reliable Node.js hosting starts at $3/month, and $1/month static HTML hosting

Created with love in Canada, visit hostnodejs.com today

Feel like to post an Ad? Learn Details
All Projects → usc-isi-i2 → festival-text-to-speech-service

usc-isi-i2 / festival-text-to-speech-service

Licence: Apache-2.0 license
REST service to call the Festival text to speech application

Programming Languages

C++
36643 projects - #6 most used programming language
c
50402 projects - #5 most used programming language
scheme
763 projects
java
68154 projects - #9 most used programming language
shell
77523 projects

festival-text-to-speech-service

###Overview This sample web service calls upon a Festival server to do synthesize text into audio. Communication between the browser, web service, and Festival service is as follows:

  1. Browser sends HTTP request to the Java-based web service

  2. Web service acts as a client and calls the Festival speech server

  3. Festival server returns audio as a byte array

  4. Web service handles the byte array audio and returns it to the browser

The Festival TTS server returns all audio as some form of uncompressed format such as WAV. The web service provided in this repository uses the [JAVE] (http://www.sauronsoftware.it/projects/jave/) library to convert into MP3.

###Installation Instructions for Festival ####Requirements

  • Unix-based environment, preferably Linux (Do NOT use Windows with Cygwin)
  • A C++ compiler
  • GNU Make any recent version
  • NCurses library

####Installing the ncurses library

  • CentOS users can run a sudo yum install ncurses-devel to get the ncurses library.
  • Ubuntu users use sudo apt-get install libncurses5-dev

####Deciding which to version to use

  • As of December 2014, the newest version of Festival is 2.4. This version of festival and the voices included with this distribution are the highest quality so far. Because the voice synthesis engines change in this distribution, it is only compatible with the voices included in its download package. Similarly, these new voices are not compatible with older versions of festival. Version 2.4 has the highest quality of voices, but at a significant speed cost. For example, the sentence "Pablo Ruiz y Picasso, also known as Pablo Picasso, was a Spanish painter, sculptor, printmaker, ceramicist, stage designer, poet and playwright who spent most of his adult life in France." takes about 30 seconds to synthesize, while it only takes 7 seconds to synthesize in version 2.1. Instructions and recommendations for older releases and additional voices for these voices are located on the bottom of the README.

####Installing Festival 2.4

  1. Go to [http://www.festvox.org/packed/festival/2.4/] (http://www.festvox.org/packed/festival/2.4/) and download festival-2.4-release.tar.gz, festlex_CMU.tar.gz, festlex_OALD.tar.gz, festlex_POSLEX.tar.gz, and speech_tools-2.4-release.tar.gz. Since a lot of the old voices do not work with this new version of festival, you should download at least one voice from the voices/ directory.

  2. Unzip all the archives using a "tar xzf " command or "for a in ls -1 *.tar.gz; do tar -zxvf $a; done" and move the "festival" and "speech-tools" folders into whatever directory you wish Festival to rest in

  3. Change directories to wherever festival is

cd speech-tools
./configure
make
make test
make install
cd ../festival
./configure
make
make test #optional
make install

###Running the Festival server

  • To run the festival server, either export the PATH to point to the bin inside the festival folder (export PATH={path to main festival folder}/festival/bin:$PATH), or manually type out the path and do a festival --server from the command line. Of course, include the "./" prefix if done without exporting the PATH.
  • By default, Festival runs on port 1314. To change the port at runtime, add "'(set! server_port PORT)'" to the command, where PORT is the desired port number. Ex: festival --server '(set! server_port 1515)'
  • To edit any defaults, including the port numbers, number of concurrent clients, security settings, etc...Edit the festival.scm file (located at festival>lib). Information about the server settings begins at around line 408.
  • An init script can be used to run festival as a service and enable automatic startup. An example script for CentOS has been provided. Ubuntu users may refer to [https://github.com/zeehio/festival-debian/blob/master/debian/festival.init] (https://github.com/zeehio/festival-debian/blob/master/debian/festival.init).
  • To change the default word parsing rules (such as forcing festival to pronounce & as "and", for example), one can edit the token.scm file (located at festival>lib). The one updloaded here has 2 added rules: add silent vowels to 3+ letter words so that festival will attempt to pronouce them as opposed to spelling them out, and to replace & with "and".

###Additional information

###Creating the web service (using Maven)

  • An example web service is included in the "festival_service" folder using the Jersey framework. The service has dependencies on the JAVE library as well as the Jersey framework. A local repository has already been included (repo/) with JAVE installed in it. Jersey can normally be downloaded from the main maven repository, though Maven 2.2.1 fails to follow the 301 redirect requests, leading to build errors. This can be bypassed by either downgrading to 2.2.0 or upgrading to 3.x. Alternatively, one can try to manually install the jar files (located inside the lib/ folder) using commands of the following format:
mvn deploy:deploy-file -Durl=file:repo/ -Dfile=lib/{name-of-jar-file} -DgroupId={groupId} -DartifactId={artifactId} -Dpackaging=jar -Dversion={version #}

Once all dependencies are taken care of, the project can be compiled using

mvn clean install

This will generate a .war file. If you don't wish to change the code, then using the provided .war file is fine.

  • Details about the source code:
    • Client.java is the actual Festival client. It communicates with the Festival server using Scheme commands and returns a byte array of the audio file in some uncompressed format (wav, nist, ulaw, etc.). This class handles customization of server location, voice selection, text mode (not demonstrated...see manual), and file type. Currently, the default settings for audio are WAV format and the cmu_us_slg_cg voice. Overloaded functions have been provided to change those parameters.
    • Festival_service.java demonstrates how to call the Festival client. The first function, returnSound() is very basic and gets the desired text from the URL and returns the audio file in raw, binary format. The returnFull() function receives and returns things in JSON format. It offers much more functionality, including an example of how to parse the wav-format header for information about the audio (sampling rate, number of samples). The audio file itself is returned as a base64 string inside the JSON. returnDemo() is similar to the returnSound() function, except that the voice is included as a parameter in the URL. This can be used as a simplistic way of switching between voices, without the hassle of generating a JSON query for the returnFull() function. The convertToMp3() function encodes the wav file generated by Festival, and encodes it into mp3 format using the JAVE library. Temporary WAV and MP3 files are created and deleted during the process.

###Instructions and recommendations for older versions of festival ####Version 1.96 or Version 2.1?

  • Default voices for Festival are generally low quality, so additional voices must be added. There are a higher-quality experimental multisyn voices found among the 1.96 release.
  • Among the most popular additional voices for Festival are the Nitech HTS voices and the enhanced CMU Arctic voices
  • Nitech HTS voices only work on version 1.96 of Festival, which has a harder installation process. There is, however, a single compiled HTS (female) voice available for the 2.1 release; it is found on the festival site with the other 2.1 packages
  • CMU Arctic voices and the experimental multisyn voices work on both the 1.96 and newest 2.1 versions of Festival

####Instructions for installing v1.96

  1. Download "festival.tar.gz" and "speech-tools.tar.gz" from the "Festival-1.96-revisions" folder

  2. Move these archives into whatever folder you wish Festival to rest in

tar xzf festival.tar.gz
tar xzf speech-tools.tar.gz
cd speech-tools
./configure
make
make test
make install
cd ../festival
./configure
make
make test #optional
make install

####Instructions for installing v2.1

  1. If on Ubuntu, the easiest method is to run "sudo apt-get install festival" to get the Debian version of festival. Otherwise, continue reading for how to install festival from source

  2. Go to [http://festvox.org/packed/festival/2.1/] (http://festvox.org/packed/festival/2.1/) and download festival-2.1-release.tar.gz, festlex_CMU.tar.gz, festlex_OALD.tar.gz, festlex_POSLEX.tar.gz, and speech_tools-2.1-release.tar.gz. The other archives are optional, and merely offer a set of working voices. If you wish to test the festival installation using the given test suite, download the other the remaining files as well.

  3. Unzip all the archives using a "tar xzf " command and move the "festival" and "speech-tools" folders into whatever directory you wish Festival to rest in

  4. Change directories to wherever festival is

cd speech-tools
./configure
make
make test
make install
cd ../festival
./configure
make
make test #optional
make install

####Installing Additional Voices

  • Just follow the instructions here.
  • The experimenal multisyn [here] (http://www.festvox.org/packed/festival/1.96/). Though released with the 1.96 version, they also work with v2.1. The multisyn voices should be moved into a new folder: festival>lib>voices-multisyn folder.
  • Highly recommended are the Enhanced Arctic CMU voices and/or the Nitech HTS voices. CMU voices are far bigger, but work on v2.1, while Nitech only works on v1.96.
  • For 2.1 users, a packaged female voice (cmu_us_slt_arctic_hts) that is similar to its nitech equivalent may be found on the [festival site] (http://www.festvox.org/packed/festival/2.1/). There are also two additional cg voices on that site (all should be moved to the festival>lib>voices>us folder).
  • After installing the voices, be sure to edit the voices.scm file (located in festival>lib) to include the names of whatever voices you've added. This edit should be done in the "(defvar default-voice-priority-list " function, located around line 325.
  • To make a voice the default voice, either edit the voice list voices.scm so that the desired voice is first on the list, or edit the festival.scm file (located in festival>lib) so that "(set! voice_default 'voice_{NAME OF VOICE}')" is added at the end.
  • NOTE The Enhanced Arctic CMU voices CANNOT be set as defaults for some reason. Therefore, to make sure festival is able to run, make sure to have an installed and useable voice (the default diphone voices, or even the Nitech voices) at the top of the voice list. Do not edit the festival.scm file to change the default voice.
  • Demos of the CMU and Nitech voices can be found in the Voice_demos folder. The name of the file is the full name of the voice, so this is the parameter that should be used in the web service if any voice selections are chosen.
Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].