Playback cannot be stopped using mbrola voices

[GoogleCodeExporter]

While stopping the playback using an eSpeak voice works well, you cannot stop 
the playback when using an mbrola voice – the Gespeaker application just 
»freezes« (and even turns »dim« if the text is longer) until the text was 
spoken completely.

What steps will reproduce the problem?
1. Paste a text of at least 30 words into the text area.
2. Choose an mbrola voice, e.g. »german-mbrola-6«.
3. Click on »Play« and try to cancel playback while it speaks
   by clicking on »Stop« – you will notice that the »Play«
   button doesn't even change its caption to »Stop«.

What is the expected output? What do I see instead?
   I'd have expected the playback to be stoppable – just like
   it is stoppable at all times when using an eSpeak voice,
   e.g. »german«. But instead, the Gespeaker application
   freezes (hangs) and turns dim after a while, still speaking.

What versions do I use? On what operating system?
 - Ubuntu 10.10 64-Bit on Pentium-4 DualCore @ 3.46 GHz, 2 GB RAM.
 - Linux Kernel 2.6.35-22-generic #35-Ubuntu (x86_64).
 - espeak, mbrola, gespeaker … installed via Synaptic from standard
   repositories:
   * espeak 1.44.05~really-1.44.04-0ubuntu1
   * mbrola 3.01h-5
   * gespeaker 0.7-1

Thanks in advance for checking this out, fixing this annoying bug, pointing to 
the problem or providing a workaround. Please let me know, if you need more 
information.

Regards,

Dennis

Original issue reported on code.google.com by thecondo...@googlemail.com on 6 Nov 2010 at 4:41

[GoogleCodeExporter]

Hi!

Yes, I can confirm this; using ubuntu 10.10, on a 32-bit installation. amd-quad 
core machine. I am using gespeaker-0.8.1 from this site, same espeak and mbrola 
versions as Dennis.

so long
hank  

Original comment by hanksch...@googlemail.com on 7 Nov 2010 at 4:19

[GoogleCodeExporter]

Thanks hank for the confirmation,

as you are using the 0.8.1 version of Gespeaker from this site and having the 
same problem, I assume the developers/bug maintainers can change the status of 
this bug fromn »New« to »Confirmed« or something similar now.

Regards,

Dennis

Original comment by thecondo...@googlemail.com on 7 Nov 2010 at 4:31

[GoogleCodeExporter]

I'm still using Ubuntu Karmic 9.10 and have the same problem. Used to work fine 
though until I recently built/updated espeak to version 1.44.05. The newer 
version of espeak has native access to mbrola. It used to be necessary to pipe 
espeak to mbrola. 

I looked at the python code for gespeaker some and I'm pretty sure the problem 
lies in this area. I don't understand python well enough though to make the 
changes though. 

It looks like mbrola support was added to espeak in version 1.44.01. Suspect 
some sort of checking for espeak versions and changing program flow in 
gespeaker is needed. Or you might be able to back up to an older version of 
espeak if it bothers you a lot...

Here are the pertinent parts from the espeak changelog:

espeak 1.44.05

Fix error in big-endian data conversion program, producing bad data.

Make geminated voiced stops (eg. [bb] ) longer at fast speeds.

Provide conditional compilation of the mbrola interface, define macro 
INCLUDE_MBROLA in speech.h

Mbrola: also look for mbrola voices in /usr/share/mbrola/voices

Pad TUNES and frame_t structures to a multiple of 4 bytes.

lang=da, Don't weaken unvoiced stops before pause.
lang=el, Remove final unstressed [a] if the next word starts with [a].
lang=pt, Change final [U] to [w] if next word starts with a vowel.

espeak 1.44.03

Fixes:
Lang=el, mk.  Was speaking words as individual letters.
Lang=pl.  Fix prounciation of 'ć' and 'ci'.
Fix crash in big-endian data conversion program.
Fix problem where changing voices reduces the speaking rate, at fast rates.

speak_lib.h:  add macro definitions for minimum, maximum, and normal speaking 
rate values.

espeak 1.44.01

Fix crash with very long numbers.
Speak very long numbers as individual digits.

Unpronouncable word check:
  Rules for unpronouncable initial letter sequences can now be defined in *_rules files.

  The unpronouncable word check now stops when an apostrophe is found.

Phoneme definitions:
  Optional second parameter to FMT() statement specifies a percentage amplitude.

  Added "ipa" statement to specify the IPA name for a phoneme if the default translation is not correct.

  Add phoneme "equivalents" tables, so that words can be spoken with foreign (eg English) prounuciation rules, but using local phonemes.

  New attributes:  flag1, flag2, flag3

  New attribute: nopause. Prevents the insertion of a short pause when this phoneme starts a word which follows a vowel.

  New conditions: isFlag1, isFlag2, isFlag3, isSibilant.

  New statement: InsertPhoneme()

Phonemes: improve syllablic [m-] [n-] [N-]

Mbrola:
  Command-line espeak and the libespeak library now call the mbrola program directly, rather than producing phoneme text which must be piped into mbrola.

  Added --pho command-line option to generate mbrola phoneme information (.pho data).

Phoneme output:
  Add --ipa command-line option to produce phonemes names using the International Phonetic Alphabet.

  Indicate language changes during phoneme output with: (en)  (fr)  etc.

-X command-line option: Show the matching of multiple-word entries in *_list 
files.

Speak sequences of letters and dots as individual letters and don't speak 'dot' 
(eg. "u.s.a").

Don't speak punctuation characters inside <audio> ... <audio/>.

Don't speak "dot" if an ellipsis is followed by a dot.

Vowelcharts:  Show the positions for multiple FMT() statements in a vowel 
phoneme definition.

*_rules: add attributes  $p_alt $p_alt2 $p_alt3,  $w_alt $w_alt2 $w_alt3

*_list:  add attributes:  $sentence, $atstart

klatt synthesizer: implement echo (defined in voice files).

espeakedit:
  Prosody display: Show stressed and secondary-stress syllables.

  Remember window size and position.

  Change the frame-length field from Spin Control to Text Control to allow better access from screen-readers.

Intonation:
  New file, 'phsource/intonation' to define 'tunes' which can be used from voice files.

  espeakedit: add Compile -> Intonation data

  Intonation: change the internal pitch unit to give finer control, and align with the values displayed in the espeakedit Prosody window.

Speed: Increase range to 80 to 450, with default=175.
Improve speaking at high speeds.

Language options: add an option to the Regressive Voicing option to de-voice 
the final consonant of words.

lang=ta, hi.  Letter-names for combining vowel characters are distinguished 
from stand-alone vowel characters by adding an initial click sound.

lang=en: Reduce consecutive unstressed syllables to 'diminished' stress, only 
in unstressed words.
lang=de: Change 'r' phoneme.
lang=es: Improve the rules for reducing 'b', 'd', 'g' to approximants [B] {D] 
[Q].
Language improvements include: Danish, Dutch.

espeak 1.43.03  (bug fixes)

Fix crash when embedded control codes are followed by numbers of 5 or more 
digits.
Fix lang=hu,  First character of an abbreviation is missed after an ordinal 
number (eg."2. cd")
Fix XML tag not recognized after "..." when announce punctuation is enabled.
Fix lang=zh-yue, 'p' 't' 'k' after a vowel give a long pause.
Fix lang=ru, "o" missing in unstressed syllables.

espeak 1.43.02

Language improvements including Danish.
Fix: "  50000" with leading spaces was spoken as "50".
Don't consider multiple spaces as a thousands separator (eg. "2  000").
Fixed phoneme [n^] for klatt synthesizer.
Lang=Hungarian, don't allow dot as thousands separator.

espeak 1.43...

Original comment by lf...@iserv.net on 10 Nov 2010 at 5:54

[GoogleCodeExporter]

could you please provide a sample text which produces the defect?

Original comment by mufl...@vbsimple.net on 13 Nov 2010 at 7:35

[GoogleCodeExporter]

Hi!

Any german file will do; I'll append one. (Actually I just noticed that 
Gespaker will not open all text files; I just tried one produced by ocre, and 
it didn't work out -- but thats a different story...) 

The feezing occures whenever a text is read using (german) mbrola voices, a 
text written, or one imported via open -> file.

I didn't check with other languages, though.

(OT: Glad you liked my translation/po file ;-))

so long
hank 

Original comment by hanksch...@googlemail.com on 13 Nov 2010 at 9:04

Attachments:

• hannah006.txt

[GoogleCodeExporter]

E.g. try this:

Dies ist ein Test. Dies ist ein Test. Dies ist ein Test. Dies ist ein Test. 
Dies ist ein Test. Dies ist ein Test. Dies ist ein Test. Dies ist ein Test. 
Dies ist ein Test. Dies ist ein Test. Ende des Tests.

Langage: german-mbrola-6
  Pitch: 50
 Volume: 200
  Speed: 160
  Delay: 5

Then click on play. Try to stop playback while listening. The Gespeaker 
application freezes until ALL words are spoken.

Original comment by thecondo...@googlemail.com on 13 Nov 2010 at 9:22

[GoogleCodeExporter]

I'm unable to reproduce it with my machine using both texts and both 
german-mbrola-6 and german-mbrola-7.

I'll try againt with some other machines

Original comment by mufl...@vbsimple.net on 14 Nov 2010 at 9:33

• Added labels: Priority-High, Usability
• Removed labels: Priority-Medium

[GoogleCodeExporter]

problem acknowledged

This is an issue related to espeak 1.44 (1.43.48) which doesn't write phonemes 
anymore to stdout to let mbrola to play them as indicated in the official 
website:
http://espeak.sourceforge.net/test/latest.html
http://espeak.sourceforge.net/mbrola.html

the new espeak now plays the audio on by itself, breaking the compatibility 
with older softwares.

A temporary fix could be editing this line in 
/usr/share/gespeaker/src/Settings.py
argsEspeak = '-a %v -p %p -s %s -g %d -v %l -f %f --pho'

I'll fix it in the next release

Thank you for your report

Original comment by mufl...@vbsimple.net on 14 Nov 2010 at 11:06

• Changed state: Started
• Added labels: Milestone-NextVersion

[GoogleCodeExporter]

Hi!

Thanks for the fix - it works fine! :-)

so long
hank

Original comment by hanksch...@googlemail.com on 14 Nov 2010 at 12:10