ECE 191 Group 6 Spring 2006 Embedded Speech Recognition System

1
ECE 191 - Group 6Spring 2006Embedded Speech
Recognition System

• Group Participants Adam Ansari, Abdifatah
  Hagisufi, Chieh-Wei Huang, David Wu, Weixuan He
• Advisors Rajesh Hedge, Bhaskar Rao
• Sponsor CaliT2

2
Agenda

• Gantt Chart
• Tasks
• Technical content
• Summary
• Plans for Next Week

3
(No Transcript)
4
Software

• Embedded visual C create and develop windows
  CE-based applications
• Pocket pc emulator simulates a Windows-CE based
  platform. You can design and build a Windows
  CEbased platform and test it using software that
  mimics hardware rather than testing the platform
  on hardware.

5
Embedded Visual C 3.0
6
Embedded Visual C 4.0
7
ActiveSync

• - Gateway between the Windows powered PC and
  Windows Mobile powered device
• - Great synchronization experience with
  Windows-based PCs and Microsoft Outlook.
• - Can synchronize directly with Microsoft
  Exchange Server 2003.

8
HMMs in Depth

• Hidden Markov Models run under two assumptions
• 1. Speech may be split into segments, states, in
  which the speech waveform may be considered
  stationary. The transition between these states
  is assumed to be instantaneous.
• 2. The probability of a certain observation being
  generated is only dependent on the current state.

9

• Hidden Markov Model is a finite state machine
  that generates random observation vectors.
• Each state contains observed Gaussian
  distributions from a data base trained by user.
• The word hidden refers to the transition
  probabilities and the most likely state sequence
  for a given model.
• The object is to maximize the probability of the
  observation vectors given a specific Markov model

10
(No Transcript)
11
Summary

• We had a little trouble using the emulator, but
  we have figured out the reason for errors.
• Finished up research on Hidden Markov Models

12
Plan for Next Week

• Make an interface code using C for the speech
  recognition engine
• Individual enrollment
• Testing