Maria-Mirela Petrea, Dan Cristea * Dealingwith Prosody. A Computer-Assisted Language Learning Approach




2. The tool

PROSODICS is an application intended to assist people to improve their abilities to produce utterances in a foreign language as close as possible to a native speaker pronunciation. The package is addressed to two levels of users: STUDENT and MASTER. The main user of PROSODICS is the STUDENT. To use the application, she should possess the minimum computer expertise that is usually required when dealing with Macintosh style menus and windows. The kinds of actions the STUDENT will have to do are the following: to listen to the master voice speaking the utterance currently displayed, to record her/his voice when uttering the same expression, to observe the comparison window mainly informing about the stress she/he puts on different segments of the uttered text matched against the ones of the master, and to repeat the last two steps until getting a graphical response reasonably close to the original (master's) one. The MASTER is the person responsible for filling up the original utterances into the computer and for editing them adequately. To save the STUDENT user from acting erroneously, the actions to which the student is allowed to are fewer than those of the master's. In this section we shall briefly describe the way PROSODICS answers to different actions, for more details the reader being referred to [10].

The File menu groups altogether the usual file operations, opening and saving files, printing the content of the current window, closing the window currently active, and quitting the application. There are two types of files a student can open, a MASTER file or a STUDENT file previously recorded; the student can also record herself and save the signal in the active STUDENT window, but cannot modify any MASTER signal, nor can record and save her signals as MASTER files. Apart from the Prosodics file format, which is extremely close to Sound Manager format 1, with some additional information about signal's amplitude range, utterance type (lexical or sentential), speaker affiliation to male or female classes, PROSODICS can access also files in Signalize format, AIFF format and can extract sounds from movies.

Each time a record action is performed, there are two files to be attached to the current signal: the proper signal files that contain the data of the speech signal, and the event files that contain data used by the program to record features and parameters of the processed signal. Therefore, two files will be produced on the disk, each sharing the name the user gave, but also having as extensions sig or eve respectively. Two filesinstead of one was preferred for testing and debugging reasons. Any signal with events attached will save computational effort next time it will be opened. There are three main windows in PROSODICS which can be accessed from Windows menu, Master signal, Student signal and Prosodics. The first and the second are quite alike, and display the wave form signal, the pitch, the loudness and the ZCR functions. The signal can be stretched/squeezed and scrolled left/right. The user can play the whole signal or a part of it, by selection. At desire, there can be displayed also markers delimiting silence, fricatives' portions, the results of the segmentation and the final speech units' delimiters resulted from signal's editing. Prosodics window is the one to which the student is most concerned with. It displays the intonational contour by means of an approximation of pitch movements and contains results of the master-student comparison, as figure 12 shows. The menu Options, as far as the student is concerned with, lets her change male/female initial setting, to display or hide segmentation markers, to see the current selection of the signal, or return to the display of the whole signal in the window, to set the type of alignment used.

Besides the actions a student is allowed to, the master can record and save MASTER files and can edit her signals (by accessing the Edit Master item in the Edit menu), that is, can adorn the signal with improved delimiters marking speech units boundaries, moreover can label segments with text information (see figure 2). The editing can be seen as an alignment operation between the speech signal and its orthographic transcription. After the actual recording has been done, the master will be shown a window where the waveform of the signal is displayed. It is crossed over by vertical lines giving moments on the time axis where the program automatically detected the most probable boundaries between speech segments. The meaning is that somewhere between these boundaries, significant distinct uttered segments of text must lie. Sometimes segments smaller than required are found by the program. Consequently, master's job will be to listen to these segments, to assemble them up into minimal length segments that belong to the same unit, and to associate to them the corresponding character string.

To sum up, the scenario PROSODICS proposes is this: a student should open a master file previously edited, listen to it, record herself and carefully watch Prosodics window. If the result is not satisfactory, she should resume.



152

Previous Index Next