Transcriber is a tool for assisting the manual annotation of speech signals.
It is more specifically designed for the annotation of broadcast news recordings, for creating corpora used in the development of automatic broadcast news transcription systems, but its features might be found useful in other areas of speech research.
Transcriber is developed with the scripting language Tcl/Tk and C extensions. It relies on the Snack sound extension, which allows support for most common audio formats, and on the tcLex lexer generator.
Transcriber is distributed as a free software under GNU General Public License. and has been built to support different platforms (Windows XP/2k, Mac OS X and various Linux).
Here are some key features of "Transcriber":
· management of several layers of segmentation: basic segmentation for orthographic transcription (e.g. at each sentence or at each breathing), speech turn segmentation (new speaker) and section segmentation (new topic...); these 3 layers are hierarchically embeded. A 4th segmentation in background acoustic conditions is also available
· annotation of named entities
· display of segmentations under the signal and in the text editor; display of each segmentation under the signal can be toggled with a contextual menu
· always synchronize the cursors in the text editor window and in the signal window: as soon as the cursor moves in one window, it is moved in the other one and appears within the same segment
· easy creation of the basic segmentation by the way of successive insertions of boundaries at current cursor position, even during playback
· move segment boundaries with the mouse
· insert mark of change in background acoustic condition (music, speech, other) at current cursor position in the signal
· indicate some boundaries as the beginning of a new speech turn or of a new section
· edition of turns, sections or background conditions with a mouse clic on the corresponding button in the editor, including possible destruction
· management of a speaker list with a precise description of each (name, type, accent...), further modification of its characteristics, research of turns involving a specific speaker, importation of speakers from another transcription
· in the same way, management of a list of topics, find sections about a topic, import topics from another file
· specific management of overlapping speech with marks for each speaker in the text editor
· insert predefined events (noise, pronouciation mark, lexical mark, language change) which can be instantaneous or be at the beginning or the end of the event; user defined display of events (font, color, format); edit or remove with a single clic on the event in the text
· standard text edition features: cut/copy/paste, find/replace, (very limited) undo
· glossary for predefined words or expressions
· transcriptions in XMl format for easier automatic processing and exchange; syntaxic validation of files upon their DTD
· user-chosen encoding is used for transcriptions input/output; most current encodings (iSO-8859-*, EuC-JP...) and unicode (uTF-8) can be used
· when opening a file, automatic research of corresponding sound file; if none is found, user is being asked to locate it.
· import .typ transcriptions in .typ format and various labelling (xwaves, OGi...); export to .typ and .stm
· periodic automatic backup
· keep previous version in backup file
· read most standards audio formats (thanks to Snack library)
· handle long duration audio files (up to several hours) - for duration over 30 sec., a low-resolution view of the signal is computed in background and stored for speeding up later display
· direct access to any position in the signal; selection of part of the signal
· synchronized cursor during playback
· several playback modes through keyboard combinations: play/pause, play the current segment, the selection, around the cursor
· interactive change in resolution without stopping playback
· able to display two views of the signal at different resolutions
· vertical zoom
· display cursor position (also during playback), selection width and other messages under the signal
· monitoring of user's productivity during the session
· information window about the signal (filename, duration, format), about the transcription (number of segments, of words...) et optionnaly of the amount of work in the current session
· edit the list of predefined events (noise, pronounciation...)
· edit keyboard shortcuts for insertion of characters or strings (including events)
· font and colors user editable
· easy localization in any language (English and French provided)
· possible backup of the current state in a configuration file for use in the next session: open files, options, window position
What's New in This Release:
· Upper extensions are now recognized as valid extensions for transcription and audio files.
· A DTD different from trans13.dtd or trans14.dtd is not considered anymore as a former DTD.
· All unicode encoding are now supported on Windows
· Transcriber 1.5.0 refused to be launched if it found an external speaker database. That file was created when the speaker turns were modified. That bug has been corrected.
· The default browser (used for viewing help files) could not be set. This bug has been corrected.