Telephone bandwith speech database: creation, applications and experiences for Polish Language

Slawomir KULA

Abstract
The paper covers challenges and experiences related to constructing and apllying a telephone speech database bandwidth in telecommunications domain in Poland. Speech processing technologies like text to speech, speech and speaker recognition or speech coding are used in many telecommunications services, especially in wireless telephony, intelligent network and computer telephony. Performance evaluation of telecommunications speech processing systems very often need a specific kind of speech database, not only lexically balanced but also containing telephone speech signals of different quality. The sources of noise and speech signal distortion in telephone chain are described. The connection from the user telephone to the network varies widely: analog connections of varying quality, ISDN, analog cellular, and three types of digital cellular are all possible. Network transmission can be analog of varying quality or digital, which greatly enhances speech quality but introduces some pitfalls. There are some differences between telecommunications networks in Western and Central and Eastern Europe. The average subscriber loop in Central and Eastern Europe is longer than in Western Europe. Due to this fact, the attenuation of speech signal is stronger. Also, transmission networks are mainly analog. A method of constructing a telephone speech database tailored to specific parameters of Polish telecommunications networks is presented.

Intelligibility evaluation results of encoded speech in cellular telephony based on a phonetically balanced database of Polish lagatoms are given. The database has been developed for this purpose at the Institute of Telecommunications, WUT. Speech transmission for the experiments was achieved by creating different kinds of telephone connections, analog and digital.

Back to Programme ml>