WWWTranscribe

WWWTranscribe is a tool for the annotation of audio signals via the WWW. It features an oscillogram display of the speech signal, audio output, editing buttons that simplify the task of annotating the signal, and a formal consistency checker for the annotations. WWWTranscribe was developed at the Bavarian Archive for Speech Signals (BAS)^8.20 within the SpeechDat project. Currently^8.21, it supports orthographic transcriptions according to the SpeechDat guidelines; other annotation systems can be added simply by extending the annotation object class hierarchy.
WWWTranscribe is implemented in Java using only the standard JDK classes to guarantee platform independence.

Figure 8.1: Transcription page of WWWTranscribe

In WWWTranscribe, the transcriber logs in and enters the ID of the session to be transcribed. A session consists of a number of recordings, each containing a single utterance corresponding to a prompt in the interview. Once a recording is selected, the transcription page is displayed. It contains a single output button with a speaker icon, a signal display, transcription and comment text fields, an assessment menu, and save and clear buttons (see figure ). A click on the speaker button outputs the speech signal as sound. For read items, the original text of the prompt sheet is displayed in the transcription field, for spontaneous speech this field is initially empty. Any text in the transcription field can be edited. The buttons below the transcription field perform some basic conversion tasks on the text in the transcription field, e.g.:

• text to lower or upper case
• digit sequences to orthographic digit or number strings
• money amounts and date expressions to orthographic strings

The assessment popup-menu allows the transcriber to select general noise markers. Comments on the recording, e.g. on the quality of the speech or the signal, may be entered into the comment field. The save button saves the transcription to the file system at the server site in the SpeechDat SAM database exchange format.

WWWTranscribe performs an automatic consistency check on the annotation text so that only formally valid annotations are entered into the annotation database.

At the BAS WWWTranscribe has been successfully used for a wide range of transcription, tagging, validation and evaluation tasks. WWWTranscribe is currently being packaged for public distribution^8.22.