The NYNEX Automated Customer Name and Address (ACNA) service trial was designed to evaluate the feasibility of automating reverse directory assistance using text-to-speech synthesis. In a reverse directory service, callers provide a telephone number and in return are given the associated name and address (listing) information. The ACNA trial automated this process, using text-to-speech synthesis to deliver the listing information. Five high-end text-to-speech devices were evaluated over the course of the experiment, with approximately 5000 real customers processed through the ACNA position. Data such as requests for repetition, spelling, and survey responses were collected to compare the participating devices. ATT's TTS was rated by callers as the most understandable device. Berkeley Speech Technology's BeST device and ATT's TTS produced the highest proportion of listings that did not elicit clarification requests. Results along several dimensions suggest that automated CNA is acceptable to customers, although it does not perform as well as operator-handled CNA. Marked differences among the high-end synthesizers tested here highlight the importance of careful evaluation, even for services as constrained as ACNA.

Keywords: evaluation methodology, text-to-speech synthesis, telephone application