|
The primary objective of speech recognition is to enable easy access to the full range of computer services and communication systems, without having to type or even to use our hands. By using a client/server approach, combined with the latest recognition systems, Distributed Speech Recognition (DSR) delivers high performance levels and access flexibility, in a way that is both practicable and affordable.
The objective of the DSR project in IBM Haifa Labs is to provide a client/server standard-based DSR software framework for the deployment of speech-enabled applications and services over 2.5G and 3G mobile networks.
The Haifa Research Lab's contribution to the new extended ETSI DSR standard enables the server to reconstruct the speech for playback. This speech reconstruction and playback option was unavailable using previous DSR standards. The additional functionality is achieved by extracting tonal and voicing information at the mobile device and transmitting it to the server alongside the spectral features. The tonal information can also be used for enhancement of tonal languages recognition. The new extensions to the ETSI DSR standards offer added capability and open the doors to new possibilities in many areas including speech reconstruction, recognition of tonal languages, telemetric, and even Multimodal applications. Since February 2004 the extended DSR ETSI standard is recommended also by 3GPP as the preferred coder for speech-enabled services.
Haifa Research Lab has demonstrated the capabilities of the DSR technology by building a client-server prototype. This system is composed of a wireless device, such as iPaq with a wireless LAN adapter or a cellular phone; and a server that provides the application and the speech recognition capabilities based on the DSR parameters that are received from the client.
Materials
| 
