At the start of each recording session, the speaker was given some instructions to read. These instructions provided a general introduction to the project and a description of the task (reading US American English business newspaper sentences to a microphone). The speakers were asked to read each sentence through before recording it. They were also warned of the possibility of grammatical and spelling errors occurring, the financial business jargon and the written language style. The speakers were instructed to speak naturally and clearly in their usual accent at normal volume. Each speaker was given a short demonstration of how to use the recording setup.
Figure 1: Computer display during a recording session
The actual recording process for each sentence consisted of five steps, each co-ordinated by clicking on screen buttons as shown in figure 1. First, the speaker silently read the sentence. When ready, the speaker would then click on record to start recording and then read the sentence aloud. The recording was terminated by clicking on the stop button. The user could check that the recording is satisfactory by using the play button to listen to what was recorded. If the speech was truncated or mispronounced it could be re-recorded by again using record. Speakers were advised to only utilise the play back and re-record facilities when not absolutely confident of the correctness of their utterance. When a speaker had finished with the current sentence, the continue button moved to the next sentence.
All speakers were first given the chance to practice on 4 sentences under supervision of the recording co-ordinator. Any problems detected in the process were corrected or explained. The practice phase was also used to set the recording levels. The results of the each practice phase were discarded.
In the adaptation phase all speakers first recorded 3 seconds of background noise. They then uttered 2 `phonetically rich' adaptation sentences, followed by the first 15 adaptation sentences from the WSJ0 adaptation corpus. During this phase the recording co-ordinator was always present and, where necessary, corrected speakers.
The main phase of each session consisted of the recording of either 90 training sentences or 80 test sentences. The recording co-ordinator was no longer present in the room during this period. Afterwards, a short anonymous questionnaire was completed by the speaker-the information from which can be found in the .ifo files.