Telephone access to an information database with voice recognition

Start Date: 01/02/1996
End Date: 31/05/1997
Funding: ΠΕΝΕΔ
Project Leader: Bakamidis Stylianos

The aim of the project was research for the development of a key phrase, speaker independent, voice recognition system, embedded in a noisy telephone environment. The user can directly access a database via the telephone network without the intervention of an operator.

Research focused on the improvement of algorithms related to noise reduction from voice signals and their adjustment to a telephone network environment, as well as on the connection of these algorithms to the ones that perform the end point detections of the signal. Research in this field is very important because isolated word recognition or connected speech recognition crucially depends on the effectiveness of these algorithms. However, at the moment these algorithms do not give satisfactory results in environments with a relatively high level of noise.

For demo purposes the system described above was connected with a small database. The overall system was the short term product that resulted from the programme.

In the long term, a pilot system of a database (e.g. addresses and telephone numbers of museums, organizations etc.) is to be developed. This system can be expanded to real size data bases with a huge volume of information. A richer vocabulary would be required to support the expanded system.