next up previous
Next: Segmentation/Alignment Up: COLLECTING A DIPHONE SET Previous: Preparing the audio collection

Collecting the data

\epsfbox{s.eps}

Figure 1: pointyclicky

For data collection, we use either an application called ``pointyclicky'' (Figure 1) written in Perl with Perl/Tk, or a simple shell script called ``prompt_them;'' both are in the FestVox distribution. With pointyclicky, the management of the recording session is simplified. You can double-click with the left mouse button to play the prompt (or the recording, if it exists), as well as make a selection and act over it - such as iterating over the selected files with the sequence of displaying the text, playing the prompt, pausing briefly, and then recording.

To aid in quickly visualizing what prompts have and have not yet been recorded, the color of the item's name, as well as the background in the scrolling list, changes based on state: if the prompt does not exist, the background of the cell is gray; if it does, it is white. If the recorded file exists, the text is blue; if not, it appears in red. This allows the user to quickly navigate over large lists and see what is and is not finished. Bad ``takes'' can be easily deleted, and their status is immediately reflected in the list.

The two menus, ``Action'' and ``Script,'' are user-extensible. The program looks into a hidden directory that contains two subdirectories of the same names, and these contain the various Perl programs that are executed when these are selected in from the menu. Items in the ``Action'' menu iterate over the items selected in the scrolling list on the left of the user interface in pointyclicky, whereas those in the ``Script'' menu act independently of the selection, to do things such as instantiate an example talking clock example, or invoke festival to perform the processing to complete the voice building.

One handy feature of pointyclicky is that, despite its name, one needn't use a mouse. All the functions are accessible as simple keyboard shortcuts, which allows the voice talent to cruise around over the files easily, and hear what is recorded and the prompts. When one wants to start recording whatever remains to be done, one simply uses the Edit menu to select the remainder, then selects the appropriate action - the default being to display the prompt, play the synthesized prompt, turn the background color to yellow, pause for 0.75 seconds, turn the background to red, record for the specified time, and then turn the background back to the default color. All of these commands have been used to extend the Perl language, and are documented in the FestVox Perl module available at http://festvox.org/pointyclicky/. Other actions are available by default too, and the user can simply drop Perl programs into the appropriate directories, or edit the Preferences for pointyclicky's behaviour in a dialog box.

Anecdotally, we have found that this interface helps both speed up the recording process, and reduce the number of errors. It certainly makes the recording process more pleasant and manageable.


next up previous
Next: Segmentation/Alignment Up: COLLECTING A DIPHONE SET Previous: Preparing the audio collection
Alan W Black 2000-10-10