SAKHR Automatic Speech Recognition ASR

1
SAKHR Automatic Speech Recognition (ASR)

• Wael Zohny
• Senior Programmer

2
Outline

• Sakhr ASR main features
• Sakhr ASR contexts
• Sakhr Microsoft platform for telephony
  applications development
• Sakhr ASR main functionality
• Sakhr Native products line
• Sakhr Auto-Attendant
• Large Directory Assistance application
• Appendix COM object building and using on VOS
  and Envox

3
Sakhr ASR main features

• Speaker independent Continuous speech recognition
  engine
• Zero Training Required
• Intelligent recognition context handling
  architecture
• Dynamic recognition contexts
• N-Best recognition result format
• Accuracy levels 98.5

4
Sakhr ASR main features

• Keyword Spotting
• Barge-in
• No initialization time for application
• Multiple simultaneous contexts recognition
• Entropy Based Speech/silence end-pointing
  mechanism
• Computer Telephony optimized structure
• (Custom Telephony Cards Recorder, Telephony ALAW
  format direct support, DTMF access)

5
Sakhr ASR contexts.

• Flat contexts.
• Grammar built contexts.
• Large Vocabulary Contexts (LVX).
• SAPI Contexts.

6
Sakhr Microsoft platform for telephony
applications development

• The Microsoft Speech development platform.
• SALT
• The SAPI interface
• Sakhr SAPI compliant engines

7
Sakhr Microsoft platform for telephony
applications development
8
Sakhr Microsoft platform for telephony
applications development

• ltform id"Form1" method"post" runat"server"gt
• ltspeechQA id"QA1" style"Z-INDEX 100 LEFT
  246px POSITION absolute TOP 215px"
  runat"server" OnClientComplete"OnRecoComplete"gt
  
• ltPrompt InlinePrompt"Please, Say the city name
  and state your request?"gtlt/Promptgt
• ltAnswersgt
• ltspeechAnswer SemanticItem"City
  XpathTrigger"/SML/CITY"gtlt/speechAnswergt
• ltspeechAnswer SemanticItem"Verb"
  XpathTrigger"/SML/VERB"gtlt/speechAnswergt
• ltspeechAnswer SemanticItem"Method"
  XpathTrigger"/SML/Method"gtlt/speechAnswergt
• ltspeechAnswer SemanticItem"Number"
  XpathTrigger"/SML/Number"gtlt/speechAnswergt
• ltspeechAnswer SemanticItem"Person"
  XpathTrigger"/SML/Name"gtlt/speechAnswergt
• lt/Answersgt
• ltReco InitialTimeout"30000" Lang"en-US"
  BabbleTimeout"100000" StartElement"Button1"
  EndSilence"1000" MaxTimeout"100000"
  StartEvent"onclick" StopEvent"onclick"
  StopElement"Button2" Mode"Single"gt
• ltGrammarsgt
• ltspeechGrammar Src"Dialer.grxml"gtlt/speechGra
  mmargt
• lt/Grammarsgt
• lt/Recogt
• lt/speechQAgt
• ltscript src"ClientProc.js"gtlt/scriptgt

9
Sakhr ASR main functionality ASR Interfaces

• ASR SDK was designed to make Integration easier
  than ever with more than one interface method
  including standard DLL, COM object, ActiveX, or
  SAPI interface. DLL, COM object, ActiveX, or SAPI
  interface.

10
Sakhr ASR main functionality Main ASR DLL APIs

• Channels Using
• HRESULT asrChannelInit(DWORD dwType, DWORD
  dwDefaultDeviceType, DWORD pdwChannelID)
• HRESULT asrChannelAddFile(DWORD dwChannelID,
  const char strRecFileName)
• HRESULT asrChannelAddBuffer(DWORD dwChannelID,
  void pvBuffer, DWORD dwDataBytes)
• Context Using
• HRESULT asrCtxLoad(const char strCtxFileName,
  DWORD pdwCtxID)
• HRESULT asrCtxUnload(DWORD dwCtxID)

11
Sakhr ASR main functionality Main ASR DLL APIs

• Request Using
• HRESULT ASR_API asrRequestCreate(void
  ppvRequestObj, DWORD dwChannelID)
• HRESULT ASR_API asrRequestDestroy(void
  pvRequestObj)
• HRESULT ASR_API asrRequestAddContext(void
  pvRequestObj, DWORD dwCtxID)

12
Sakhr ASR main functionality Main ASR DLL APIs

• Recognition
• HRESULT asrRecognizeFileCtx(DWORD dwChannelID,
  DWORD dwCtxID, const char strRecFileName, char
  strResult, double pdConfidenceScore, bool
  bConcatenate)
• HRESULT asrRecognizeBuffCtx(DWORD dwChannelID,
  DWORD dwCtxID, void pvRecBuffer, DWORD
  dwRecBufferLength, char strResult, double
  pdConfidenceScore, bool bConcatenate)
• HRESULT asrRecognizeDeviceCtx(DWORD dwChannelID,
  DWORD dwDevID, DWORD dwCtxID, char strResult,
  double pdConfidenceScore, bool bConcatenate)

13
Sakhr ASR main functionality Main ASR DLL APIs

• Recognition
• HRESULT asrRecognizeFileCtxAsync(void
  pvRequestObj, DWORD dwCtxID, const char
  strRecFileName, DWORD AsrCallBackFunctionSimpleRe
  sult)
• HRESULT asrRecognizeBuffCtxAsync(void
  pvRequestObj, DWORD dwCtxID, void pvRecBuffer,
  DWORD dwRecBufferLength, DWORD AsrCallBackFunction
  SimpleResult)
• HRESULT asrRecognizeDeviceCtxAsync(void
  pvRequestObj, DWORD dwDevID, DWORD dwCtxID,
  DWORD AsrCallBackFunctionSimpleResult)

14
Sakhr ASR main functionality Main ASR DLL APIs

• Results Handling
• HRESULT asrGetResultsCount(DWORD
  dwResultCollectionID, DWORD pdwResultsCount)
• HRESULT asrGetAlternativesCount(DWORD
  dwResultCollectionID, DWORD dwResultIndex, DWORD
  pAltsCount)
• HRESULT asrGetResultAt(DWORD dwResultCollectionID,
  DWORD nResultIndex, DWORD dwAltIndex, char
  strUtterence)
• HRESULT asrFilterResultUtterence(char
  strUtterenceIn, char strUtterenceOut)
• HRESULT asrMapResultUtterence(char
  strUtterenceIn, char strUtterenceOut)

15
Sakhr Native products line

• Sakhr Auto-Attendant (ALLO)
• Large Directory Assistance application
• (more than 100,000 Names)

16
Sakhr Native products line Sakhr
Auto-Attendant (ALLO)

• Incoming call management
• Outgoing call management

17
Sakhr Native products line Sakhr
Auto-Attendant (ALLO)

• Incoming call management

ASR
18
Sakhr Native products line Sakhr
Auto-Attendant (ALLO)

• Outgoing call management
• No need for human operators(24/7)
• Business and private calls classification
• Security, Privacy 
• Send Voice E-mails
• Monthly detailed bill for each employee (through
  e-mail or web)
• Administration Reports for each department, each
  employee calls cost
• Easy deployment through simple web interface

19
Sakhr Native products line Sakhr
Auto-Attendant (ALLO)
20
Sakhr Native products line Large Directory
Assistance Application

• Helps the user to retrieve the subscriber number.
• Contains White and Yellow pages.
• Superior to human operated system.
• Speech technology is a MUST.
• ASR (Automatic Speech Recognition)
• TTS (Text-To-Speech)

21
Sakhr Native products line Large Directory
Assistance Application

• Regional Limitations
• No clear last name and in many cases multiple
  last names.
• Complex names.
• Multiple entries with the same names.(Look-alikes
  )
• Many combinations alternates for the same entry.

22
Sakhr Native products line Large Directory
Assistance Application
23
Appendix COM object building and using on VOS
and Envox.

• COM Objects building
• VB
• VC
• VC
• Using in CT
• VOS
• Envox

24

• Thank you