A Speech Processing Solution in a 3G Media Server Miikka Rautapää Nokia Networks Supervisor: Professor Raimo Kantola 04.11.2003 Abbreviations AEC ALC AMR AMR-WB BSS DSP EC EFR FR MIPS NS RAN TFO TrFO Acoustic Echo Cancellation Adaptive Level Control Adaptive Multi-Rate (speech codec) Wideband AMR (speech codec) Base Station System Digital Signal Processor Echo Cancellation Enhanced Full Rate (speech codec) Full Rate (speech codec) Million Instructions Per Second Noise Suppression Radio Access Network Tandem Free Operation Transcoder Free Operation Agenda Problem Description Speech processing applications 3G Network MGW architecture Speech processing media server Testing Results Conclusions & Future Work Problem Description The problem MGW provides services for different interfaces Every service has own speech processing applications BSS Control PSTN IP RAN The objective Find a solution to share speech processing applications for all MGW terminations regardless of their interfaces Used methodology Case: own solution for the media gateway project ATM Speech Processing Applications Speech coding Tandem coding avoidance (e.g. TFO, TrFO) Speech enhancements (e.g. EC, AEC, NS, ALC) Speech synthesis Speech recognition Conferencing, Announcements, Interactive voice response Etc… Need lots of processing power 3G Network (1/2) Next-generation network Media gateway (MGW) routes traffic between packet network and circuit-switched network. Media server processes, manages, and delivers the media resources. Open, scalable architecture 3G Network (2/2) 3GPP release 4 network MGW MSC Server IP/ATM backbone Simplified architecture of the 3GPP release 4 network MGW Architecture Interfaces ATM, RAN PSTN, BSS IP DSP Unit Control processor DSP chips DSP software MGW applications provide different user plane framing protocols for different interfaces, and different supplementary services (e.g. announcements) for other MGW applications. MGW server processes provide basic functions for MGW applications. DSP Platform provides connection management, timer services and basic operating system functions. Speech Processing Media Server Integrated part of the MGW DSP Serves MGW applications with different speech processing applications FR, EFR, AMR ALC V-model design Client/server Signalling interface * Own contribution Testing Unit, integration and system testing True media gateway hardware Bit-exact tests Performance measurements Codecs Overhead (control code) Results 510 test cases (passed) Control Codecs 60 436 ALC 8 Multi channel 6 Performance measurements (encoder&decoder) FR MIPS 2.2 EFR AMR DTX Overhead 11.1 8.0-11.1 0.2 0.16-0.24 Conclusions Speech processing applications Media server Optimised as enhanced services platform Integrated to the MGW DSP Several applications Need lots of processing power Layered software (applications – servers – platform) Results 510 test cases Performance depends on application algorithms Future Work Other applications (e.g. AMR-WB, AEC) Easy to integrate Interface Signalling Function calls