History of Transcriber versions
Version 1.4 (mostly a stable version
of 1.3 beta)
Released on September 27, 1999
needs Tcl/Tk 8.0-8.2, Snack 1.6, tcLex 1.2
compiled under Linux/i86 (Debian 1.3/2.0), Solaris (SunOS 5.5), SGI (IRIX
6.2), Windows NT.
New or modified functions:
- handling of raw sound files is improved - a default header size can
be given in the [Options]/[Audio file...] window; changes
to the default raw sound files settings immediately apply to the current
sound file if necessary.
- background color of selected signal can be modified by the user - this
can be necessary when not enough colors are available in the colormap.
- a sample of speech conversation transcription is provided in the demo
sub-directory.
- the Help is viewed through 'netscape' when it is available.
- a -patch directory_name option has been added to the
command line options for 'trans'; this will allow loading some extensions
to the tool at startup in the future.
Implementation changes or bug fixes:
- no more Makefile at the root directory - all compilation and installation
is done from within the 'src' sub-directory.
- the DTD has been extended with the "elapsed_time" field in
the Trans tag (intended for total transcription time monitoring, but not
currently used)
- shape calculation mode disabled by default for Windows platform; background
calculation is always disabled.
Version 1.3b7
Released on July 6, 1999
needs Tcl/Tk 8.0 or 8.1, Snack 1.5p2 or 1.6a2 (do not use 1.6b1 under Linux),
tcLex 1.1.3 (do not use older release with Tcl/Tk8.1)
compiled under Linux/i86 (Debian 1.3/2.0), Linux/alpha, Solaris (SunOS
5.5), SGI (IRIX 6.2), Windows NT.
New or modified functions:
- check spelling (GUI for using 'ispell' external program)
- upon reading a transcription, look for a more recent automatically
saved version and ask user to restore from this version instead (in case
of crash during previous session)
- playback with pauses at segment boundaries (menu [Signal]/[Play
with pauses])
- significant update of French user and reference manuals
- remote sound file server is functional again (tcl/SoundServer.tcl has
to be launched on the host)
- menu for sorting an edited list (bindings, localization, etc.) upon
any field (name, value...)
- slight modifications in .stm format
Implementation changes or bug fixes:
- no clock calibration needed anymore (instead, native Snack "audio
elapsedTime" function available since 1.5p2 is used); cursor and callbacks
management in tcl/Play.tcl updated.
- mp3 sound files work with the most recent Snack versions (1.6a2).
- new Snack sub-command "order", to be used in conjunction
with "shape" and "get" sub-commands for remote sound
file access in heterogeneous network.
- use a specific Snack API allowing support of Transcriber under Windows
NT platform.
- updated "Help" menu for display of correct .html files.
Version 1.3b6
Released on May 17, 1999
New or modified functions:
- completely new French reference and user manual (HTML format)
- current cursor position in signal and vertical zoom are now saved with
session configuration
- at user option, saving current configuration is proposed before leaving
- direct access to a cursor position is now possible with menu [Signal]/[Go
to...] or by clicking on the message under the signal
- direct control of sound level in [Signal]/[Control Panel]; vertical
zoom moved to the same window
- at startup, sound files are automatically detected, even if their suffix
is unknown or lacking (in 1.3b5, they were rejected with the message "format
unknown"); more suffixes are now recognized (.aif, .aiff, .smp)
- move segment boundaries can also be done using control-left button
1 (instead of middle button) for 2-buttons only mouse users.
- modification date and version are not editable any more in Episode
window.
- language list is now displayed in lexicographical order of the current
langugage.
Implementation changes or bug fixes:
- corrected bug in 1.3b5 with activity logging when day changed during
session (caused infinite loop)
- corrected bug in update of turn and section button colors in the text
editor
- changed duration for recalibration of clock to 10 seconds.
To be corrected or not implemented:
- "undo" very limited
- no spelling correction, automatic completion etc. yet
- no printing
- mp3 sound files partially working, but randomly causing Transcriber
to crash
Version 1.3b5
Released on March 23, 1999
New or modified functions:
- transcription time monitoring has been rewritten and is displayed inside
the "Informations" window if a log file is given
- upon startup, user is asked either for a transcription file or for
an audio file.
Implementation changes or bug fixes:
- bug in 1.3b4 with turn attributes edition
Version 1.3b4
Released on March 16, 1999
New or modified functions:
- Cut/Copy/Paste of mixed text and events works as expected
- Bindings replacement string can include Events (through copy/paste
or directly with selection); binding key can be typed as-is in the editor
field
- Automatic space insertion at user option
- Save the geometry of some windows
- A user glossary is available
- Imported speaker can be selected; topics can also be imported the same
way. Unused topics and speakers can be removed. All these functions are
available with menu "Edit/Speakers" or "Edit/Topics"
Implementation changes or bug fixes:
- Bug correction in shape.c for 8bit wav sound files
- Format used in list edition since 1.3b3 slightly modified to allow
a generic n-fields edition instead of 2-fields only
Version 1.3b3
Released on March 7, 1999
New or modified functions:
- DTD: extended Event tag with attributes type (noise/lexical/pronounce/language)
and extent (begin/end/previous/next/instantaneous); previously defined
Events default to instantaneous noise.
- Edition of Event has been upgraded to ths new format
- Background shape computation is back at user option
- Events are displayed in the segmentation bar; they weren't any more
in 1.3b2
- Overlapping speech is now managed inside turn window
- Turns and sections can be destroyed from within their edition window
- Information window now also gives information about transcription
- Signal and transcription name are both displayed in the button bar
when necessary
- Better localization of the interface
- Added menu "File/Export/..." for explicit exportation to
non-native format; "File/Save" and "File/Save as..."
always use native XML format
- Default extension of transcription files has been changed to .trs;
.xml can still be used
- User configuration of:
- language list (in iso639 code)
- noise events list
- pronounce events list
- lexical events list
- display format of events, their color
- bindings
- localization strings in any language
Implementation changes or bug fixes:
- Switched to Snack1.5 (final), tcLex1.1 (final) and tcl8.0.5 for development
(should still work with previous 8.0.x)
- C sources: reworked shape.c for use of latest Snack function and shape
display of any type of sound file, including compressed ones.
- File format .typ has been corrected to come back to 1.2 behaviour
- Updated Transcriber URL to its new home page inside distribution and
docs
- no more patch to tk_getSaveFile lib for choosing the save format
Version 1.3b2
(not publically) released on February 12, 1999
New or modified functions:
- DTD:
- added channel attibutes in Turn tag (=telephone/studio)
- added scope attribute in Speaker tag (=local/global)
- Added "File/Import speakers..." menu: import all speakers
from a .xml file
- Automatic insertion of spaces before punctuations
- Suppressed "File/Close" menu
- Write to .stm format for further exploitation
Implementation changes or bug fixes:
- Modifs to Xml parsing libs
Version 1.3b1
Released on January 21, 1999 - test under Linux (Debian 1.3/2.0), Solaris
(SunOS 5.5), SGI (IRIX 6.2).
New or modified functions:
- a new XML format becomes the native format for the tool instead of
LDC's .typ
- validity of the transcription is checked upon the DTD (i.e. the syntax
of the format)
- general informations about the transcriber name, the emission or the
signal file are now kept along with the transcription
- when opening a transcription, the corresponding sound file is open
automatically
- background music or noise is now handled and displayed in a separate
transcription tier
- Speech or non-speech events and of comments are now handled as XML
tags rather than normal text
- management of overlapping speech has been reworked
- management of speakers and topics is easier and more powerful, and
allows global research and edition of speakers or topics
- automatic shape computation for large signals can be disabled at user
option
- loading raw files is once again a problem - but it should work even
better than before with future updates of Snack module...
- transcriber's productivity can be logged in a file at user option
Implementation changes:
- SoX libraries aren't used anymore; instead, all sound files are accessed
with Snack module; specific open_sound and play_sound command disapeared
- tcLex module from Frédéric Bonnet (a lexical analyser
generator for Tcl) is now used for the implementation of an object-oriented
tcl-only validating XML parser
- installation procedure and global tool structure has been reworked
thanks to D. van Leeuven useful hints
Version 1.2
Released on September 16, 1998 - test under Linux (Debian 1.3 and Redhat
4.3), Solaris, SGI (IRIX 6.2).
New functions:
- Signal:
- Display topic and speaker segmentation under the waveform along with
text segmentation.
- Ability to display a second view of the signal at a different scale.
- Possibility to open raw sound files.
- Contextual menus on the waveform for direct actions on the signal and
its display.
- Transcription:
- Undo/Redo, limited to segmentation boundary moves and text changes
(undo all text modifs in the current segment since cursor moved into it).
- Find and replace in the text.
- Automatic auto-save at periodic intervals.
- New XML format for transcription output (format will probably be modified
soon - see discussion about formats).
- Management of overlapping speech over one segment.
- More configurable options (fonts, colors, windows to display...). Most
user choices (including current sound file, transcription file and current
resolution display) are saved in the "~/.transcriber" user
configuration file upon request.
- Localization of the menus in english or in french as option; other
languages can be easily added.
- Help in HTML format.
Bug fixes:
- Topics were lost when file was saved as ".typ". Original
LDC format was slightly modified to include topic. Markup for overlapping
speech is now recognized.
- Default configuration file was named "~/.trancriber" (without
's'); it was corrected.
- Shortcut for 'Signal/Replay Segment' has been moved from <Alt-Tab>
to <Shift-Tab>; it could be a problem on some platforms.
- Menus 'Segmentation/Move to.../First segment' and '.../Last segment'
are now executed.
- ... plus a lot of minor changes
Implementation changes:
- Internal data management has been completely rewritten. Transcriptions
are stored in memory as a tree structured like an XML file, and are accessed
with a set of procedures in "tcl/Tags.tcl". Input file parsing
is very simplified (an XML-conformant library using tcLex extension for
Tcl written by Frédéric Bonnet is under development).
- Sound files are accessed via a new Tcl command "open_sound"
which calls sox and/or sphere libraries and creates a new "sound-file"
Tcl command - not (yet) compatible with Snack "sound" commands.
Playback with a new Tcl command "play_sound" which calls Snack
libraries. These functions are documented in the "doc" sub-directory.
- Sound shape (pre-computed at the resolution of 10 ms per pixel) are
not computed in background any more. Rather, the interface is stopped the
first time a large sound file is open during the shape computation; it
is then saved for the next time it is needed.
- Remote audio playback server is no longer available; it seemed not
to be essential, and Snack playback design is more robust. Remote audio
file server is still in the distribution, but the default behaviour is
to read audio files locally without launching the audio file server on
the local host. If needed, the file server should be configured before
launching to avoid security holes (e.g. by testing the client port or the
sound file directory).
- axis, wavfm and segmt new Tk widgets have been updated (but they are
not yet documented):
- wavfm widget now calls a "sound-file" command created with
the "open_sound" command for shape calculation rather than
reading/writing a channel (pipe or socket). Widget options "-width"
and "-height" added.
- segmt widget support one indirection level for time indices : start
and end segment times can be an index in an associative array. Time array
and list of segment variables are automatically traced, and the widget
is redisplayed if they are modified. Some minor other fixes.
- axis and segmt widgets are redrawn if named fonts are modified (allows
interactive fonts configuration).
- Distribution structure:
- the main distribution directory is now named Trans-x.y (x.y being 1.2
in this version). For simplification, the "editor" sub-directory
has been suppressed and its content ("doc", "img",
"src", "tcl") is in the first level. The "pkg"
sub-directory has been renamed "lib". The "bin" and
"obj" sub-directory are no longer used.
- all C extensions to Tcl/Tk are grouped in a single auto-loadable package
"libtrans.so"; there are no binary executable to be launched
in background any more.
- a patched version of the last distribution of SoX (v-12.14) is included
in the distribution in the "contrib" sub-directory
- html_library-0.3 from Stephen Uhler is included in the "lib"
sub-directory.
- Snack official distribution is used without any patch. The lite version
of the package is installed in the "lib" sub-directory.
- a configure procedure is available for the compilation of C sources
in the "src" sub-directory
Version 1.1
Released on June 26, 1998 - test under Linux (Debian), Solaris, SGI.
Modifications:
- SNACK 1.1 module from Kåre Sjölander is used instead of
Sox libraries for audio playback. This solution is more stable and portable
: Snack works under HP-UX, Sun Solaris, SGI IRIX, Linux, and Windows
95/NT. The module is used in a patched version.
- No modifications of the user interface, installation procedure still
rather crude.
Version 1.0
Released on May 25, 1998 - development under PC/Linux (Debian 1.3);
test under Sun Solaris.