Des Hommes de Parole ® WP Des Hommes de Parole ®
WP WP6: System Integration Building a Galaxy communicator compliant end-system
Des Hommes de Parole ® WP First Step: build One Box System with Dialogic telephony board under LINUX Software portage from NMS board to Dialogic boards under RH7.2 in centralized architecture. Record/play streaming, barge-in and speech detection Call control (loopstart) software Resolve Dialogic issues ( no call transfer for Dialogic global call control API ) Test with the SNCF railway info application as starting blocks for LIMSI co-workers Issues: Intermittent DISCONNECT event from Dialogic board Call transfer not yet available for application development …live with Dialogic board shortage
Des Hommes de Parole ® WP ( First Step continued: build One Box System with Dialogic telephony board under LINUX ) System requirements HARDWARE: * Minimum requirements for the PC: - Pentium GHz - RAM 512 Mo - Hard Disk 8 Go * Dialogic Board: - Long PCI slot for Dialogic board - Dialogic board D/41JCT-LS (4 channels) or D/120JCT-LS (12 channels) SOFTWARE: - LINUX RedHat Kernel V2.4.9 with latest update ( current ) + sources + headers - Driver Linux Streams (current ) - System Release 5.1 Dialogic (Runtime Dialogic and CSP library) (Continuous Speech Processing) - Dialogic Global Call Package 2.1
Des Hommes de Parole ® WP Context Tracking Context Tracking Hub Application Backend Application Backend Dialogue Management Dialogue Management Language Generation Frame Construction Speech Recognition (LIMSI-VECSYS) Galrecowrapper Sig2mfcc nsearch {Acoustic and Language models} Speech Recognition (LIMSI-VECSYS) Galrecowrapper Sig2mfcc nsearch {Acoustic and Language models} Audio (VECSYS) Galaudio Record Play Call Control Audio (VECSYS) Galaudio Record Play Call Control Text-to- Speech Second Step: Make LIMSI/VECSYS Services GC Compliant
Des Hommes de Parole ® WP Audio (VECSYS) Galaudio Record Audio (VECSYS) Galaudio Record Milestone 04-30: Record only Soundblaster card integration only Running on LINUX RH 7.2 Fixed record window only, no barge-in or silence detection Communicates with Speech recognition service in stream brokering mode Speech Recognition (LIMSI-VECSYS) Galrecowrapper Sig2mfcc nsearch {Acoustic and Language models} Speech Recognition (LIMSI-VECSYS) Galrecowrapper Sig2mfcc nsearch {Acoustic and Language models} Milestone 04-30: Acoustic and Language models to be provided by LIMSI/VECSYS based on data from partners and WP2 Running on LINUX RH 7.2 Communicates with Audio service in stream brokering mode Input audio stream in PCM 8KHz/16 bit Little Endian ( Second Step continued: Make VECSYS Oasis Services GC Compliant )
Des Hommes de Parole ® WP How the system works Audio Speech Recognition Speech Recognition Hub Recognizer.pgm SERVER: Audio HOST:localhost PORT:2800 OPERATIONS: reinitialize Record SERVER: recognizer HOST:localhost PORT:11000 OPERATIONS: Recognize PROGRAM: FromAudio RULE: :binary_port & :binary_host & :call_id --> Recognize IN: :binary_port :binary_host :call_id OUT: none! PROGRAM: ToAudio RULE: :string --> Audio.Record IN: :string OUT: none! reco.csh audio.csh hub.csh
Des Hommes de Parole ® WP ( How the system works continued ) Audio Speech Recognition Speech Recognition Hub provider: for localhost:2800 GalIO_CommDispatchFrame: got reply: {c Audio :signatures ( ( « Record » …) ( « reinitialize » …) ) :properties {c server_properties } :extra_service_types ( ) :protocol_version 1 }
Des Hommes de Parole ® WP ( How the system works continued ) Audio Speech Recognition Speech Recognition Hub “Recognize” provider: for localhost:11000 GalIO_CommDispatchFrame: got reply: {c Recognizer :signatures ( ( "Recognize", ) :properties {c server_properties } :extra_service_types ( ) :protocol_version 1 } {c FromAudio :binary_host " " :binary_port 2800 :call_id " :27083:0" :session_id "Default" :tidx 1 }
Des Hommes de Parole ® WP ( How the system works continued ) Audio Speech Recognition Speech Recognition Hub “Audio received” Got new message from provider for localhost:11000 {c ToAudio :notification "Audio received." :language "French" :string " Société Générale" :session_id "Default" :tidx 2 }
Des Hommes de Parole ® WP Context Tracking Context Tracking Hub Application Backend Application Backend Dialogue Management Dialogue Management Language Generation Frame Construction Speech Recognition Audio Text-to- Speech Next Steps: Galaxy compliant Telephony Audio server with barge-in capability and speech detection interaction with partner services
Des Hommes de Parole ® WP ( End-system) Integration questions How is the IVR set up to route calls before the agent picks up? (question from meeting minutes 2002_0226) Brainstorming… Different types of routing paths for internal call routing to a new dialogue application and/or external call routing through the public network (PSTN) or PBX ? Specification of call control interface ( pickup, hangup, explicit call transfer and blind call transfer ) between Dialogue manager server and Audio Server client ( Telephony IVR service ) ? => Definition of frames Configuration ( database queries ) GUI for call routing setup to live operator PBX extensions ( integration in call centers ) ? Usage of the Natural Language Generator (VECSYS/LIMSI) for French ? Which TTS server for English and German ?
Des Hommes de Parole ® WP Contact: Christoph Stier (VECSYS R&D)