A. Project Goal: Target 190 subjects, balanced by gender, under varied environmental conditions to participate in (10+) 5-6 minute conversations on GSM cellular phones. The speech data to be collected for research, development, and evaluation of automatic systems for speech-to-text conversion, talker identification, language identification and speech signal detection purposes.
B. Staff:(FT) Dave Graff, David Miller, Kevin Walker, (PT) Eleftheria Ahtarid, Suzanne Dorf, Bethany Klein, Nasrin Nourizad, Farimah Partovi.
C. Recruitment: Recruitment for SWB_CELL commenced 10/01/00. LDC advertised in several local newspapers, posted flyers and electronic announcements and recruited from within the OmniPoint community (LDC established contacts with OmniPoint because they are the only GSM service provider in the Philadelphia area). LDC's most successful means of recruitment was within the OmniPoint community. The LDC recruited 293 potential participants (from PA, NY, NJ, FL, CT and MN) for this study. 87% of SWB_Cell recruits (254) partcipated in the SWB_CELL project, 125 females and 129 males. The total number of recruits and actual participants, including gender distribution can be viewed in Figure (C1).
D. Participant Registration: Participant registration for SWB_CELL (as in all SWB projects) was/is a five step process. These steps, in which the potential participant contacted the LDC, provided demographic information, registered a handset(s), chose preferred topics and was provided with a PIN, can be viewed in Figures (D1), (D2), (D3), (D4), (D5). During the registration process the LDC acquired the demographic and cellular handset information (later augmented via participants call-backs) pertinent to this study. After registering, participants were sent a mailing with detailed instructions regarding the study and a written declaration that their calls will be recorded. This letter can be viewed in Figure (D6). If the potential participant agreed to the study's terms, they then called the LDC's registration number (1-800-380-PENN) to activate their PIN (D3 above).
E. Call Collection. SWB_CELL collection commenced 11/12/99 and finished 05/15/00. All participant payments have been sent to the University's accounts payable department for processing.
The SWB platform was in operation 7 days a week from noon to midnite. Call collection is driven largely by the availability schedules generated during participant recruitment (D4, above). The graph in Figure (E1) shows the nominal number of available participants per 1 - hour timeslot for each day of the week. The actual number of available participants is impacted by caller activity (participants who have already made a call in the past 24 hours), non-participation (participants who are deselected by failing to answer during when called), and total platform activity (non-participation will result in deselection at different rates depending on the number of attempted calls at a given time). This graph also shows the number of successful calls per timeslot (pink line).
During the collection period the LDC collected a total of 1309 calls, or 2618 sides (1957 GSM), from 254 participants, all of which have been audited using the SWA interface found in Figure (E2). During the auditing process, the auditors check for: 1. speaker identification/verification, 2. echo (presence of the other channel), 3. Background noise, 4. Line Distortion and 5. whether the call is empty. Auditors also have the option of entering comments about a particular call via the comments window. Comments are typically used for noting persistant conditions of a call or speaker. For example, if a speaker has a cold, this is noted in the comments field.
The graph found in Figure (E3) plots call collection throughout the collection period. Figure (E4) illustrates this information by gender. The collection average for the study's duration was 6.42 calls per day. SWB_CELL call collection statistics are available in (E5), (E6) (Gender Distribution by Call Plateau, Table), (E7) (Gender Distribution by by Call Plateau, Chart) and (E8) which shows the percentages of participants broken down by the number of calls made, received and total.
The average recorded call side for successful calls in SWB CELL was 358.165 seconds in duration. A histogram of this data is provided in Figure (E9). The average calls per participant was 10.34. 66% of SWB participants achieved the project's goal of (10+) calls. Furthermore, 68% of participants making (1+) calls successfully completed the study (10+calls). Based on the final call numbers for this project, SWB_CELL was 90% successful in achieving the sponsor goals for this study.
One significant problem which we encountered during collection involved participant call reception. We believe this problem was a result of participants' habit of turning off their cell phones when not using them. To counter this problem and to generally improve customer care, we inititated multiple participant call-backs and mailings. We also instituted a participant lottery (Six $500 dollar prizes for participants reaching the 10+ call plateau) during the final two weeks of April 00.
For the first time, the LDC conducted a post-study survey among participants. The results, which indicate a positive participant attitude towards the SWB study, can be viewed in Figure (E10).
The LDC will use the information gleaned in this phase of SWB_CELL (including the post-study survey) to improve our performance for the next phase of cellular call collection.