Cicero TTS: A Small, Fast and Free Text-To-Speech Engine.

Copyright 2003-2008 Nicolas Pitre <nico@cam.org>
Copyright 2003-2008 Stéphane Doyon <s.doyon@videotron.ca>

Version 0.7.2, June 2008

This software is distributed under the GNU General Public Licence (GPL). 
See the COPYING file in this package for more details.

Our TTS engine currently speaks French and some resemblance of English, 
although we hope it can some day be taught to speak good English or other 
languages. The engine uses context-sensitive rules to produce phonemes 
from the text. It relies on MBROLA 
(http://tcts.fpms.ac.be/synthesis/mbrola.html) to generate actual audio 
output from the phonemes. The TTS engine is implemented using the Python 
programming language.

We've come up with this TTS to try and meet our own needs as blind users. 
It's designed to be plugged as output to some screen-review software: 
currently only used with BRLTTY, but should adapt easily to other 
software. We favor speed and intelligibility over perfect pronunciation. 
We've tried to give it a quick response time, the ability to quickly 
shut-up and skip to another utterance, intelligibility where it counts 
(not perfect pronunciation), the ability to track speech progression, 
relative simplicity (hackability) and relative small code size.

This TTS doesn't do any sort of grammatical analysis and it doesn't yet 
have a dictionary, so pronunciation isn't always perfect. But we've 
stretched the rule system pretty far and we think it's getting good. We're 
definitely not linguists, but we are demanding blind users. We've been 
using it seriously to do actual work for more than a year now.

This is still an early release of our TTS, quality is beta-ish. 
Installation/integration surely has rough edges still, and pronunciation 
is constantly improving. The TODO-list is still generous.

Credits: We were very much inspired by previous implementations of this
general phonetization algorithm:

  David Haubensack's perl_tts:
    ftp://tcts.fpms.ac.be/pub/mbrola/tts/French/perl_tts.zip

  Alistair Conkie's perl_ttp:
    ftp://tcts.fpms.ac.be/pub/mbrola/tts/English/perl_ttp.zip

Extra Requirements:

- Requires python 2.3 or later (mainly for the ossaudiodev module),

- Requires mbrola: tested with mbrola 301h.

- Requires an mbrola voice database: tested with fr1.

- Requires a Linux system with a working sound card using either the OSS
  sound API or ALSA's OSS emulation.

Installation:

At this stage we just execute it directly in the directory containing
this README.

- Copy config.py.sample to config.py.

- Edit config.py to make the paths point to the right place.

- Test by running tts_shell.py then typing in some text.

BRLTTY usage with full text tracking:

BRLTTY home site is here: http://mielke.cc/brltty/
The "External Speech" driver from BRLTTY is used.

- Copy brl_es_wrapper.sample to brl_es_wrapper.

- Edit brl_es_wrapper to make the paths point to the right place.

- In /etc/brltty.conf put the following lines:

speech-driver es
speech-parameters program=/home/foo/cicero/brl_es_wrapper,uid=500,gid=500

  with the correct path where you've untar'ed this, and with
  your user's correct uid and gid.

- Restart BRLTTY.

- If it doesn't speak, go look into the log file that you
  specified in brl_es_wrapper.

Available utility scripts in this package:

- tts_shell.py: "Interactively" speaks text received on stdin, echoes it
  back to stdout but only as it is spoken. Empty line means shut-up.
  Use ctrl-D to quit.

- bulktalk.py: takes text from stdin (can be piped) or cmdline and
  either speaks it directly through MBROLA or saves the audio to a file.
  Matched rules can optionally be displayed for rule debugging.

- wphons.py: Translates phrase given on cmd line or stdin into phonemes.
  Shows each rule as it is applied. Helps track pronunciation rule
  glitches.

- saypho.py: Takes a sequence of phonemes on cmd line and speaks them
  to test out pronunciations. Calls mbrola and plays the result
  with sox.

- regress.py: checklist.en contains a list of phrases and known
  good phonetic pronunciation pairs. Ttranslates each phrase
  to phonemes using the current rules file and checks that the result is
  identical to the known good pronunciation. For regression testing
  when editing the rules.

- testfilters.py: Takes text from cmd line or stdin and passes it through
  the filters stage only, and prints the result.

Feedback Appreciated:

If you have any improvements to contribute, bugs to declare, or simply any 
comments (positive or negative) about this program, please don't hesitate 
to send us an email to the following addresses:

	Nicolas Pitre <nico@cam.org>
	Stéphane Doyon <s.doyon@videotron.ca>