Speech Recognition Application

1
Speech Recognition Application

• Voice Enabled Phone Directory
• - Yousef Rabah
• ???? ???? -

2
Why Speech Enabled Phone Directory

• Growing Technology
• Easy Access
• Mainly used for
• Educational purposes
• People with certain Disabilities
• Mobile use

3
Problem

• Automatic speech interacting phone directory
  assistance

4
Automatic Speech Recognition - Sphinx

• Speaker Dependent vs. Independent
• Acoustic modeling
• Isolated vs. Continuous
• HMM Probabilities, Parameters, Training
• Language Model
• Unigrams
• Bigrams P(word2 word1)
• Phonemes
• Lexicon Structure
• ZERO Z IH R OW
• TWO T UW
• H A HEIGH H

5
Input / Output

• 24003 samples in file /usr/local/share/sphinx3/mod
  el/lm/an4/hell.raw
• INFO live.c(239) live_nfeatvec 13
• INFO main_live_pretend.c(92) PARTIAL HYP
• INFO live.c(239) live_nfeatvec 12
• INFO main_live_pretend.c(92) PARTIAL HYP
  A(2)
• INFO live.c(239) live_nfeatvec 13
• INFO main_live_pretend.c(92) PARTIAL HYP
  EIGHTH
• INFO live.c(239) live_nfeatvec 12
• INFO main_live_pretend.c(92) PARTIAL HYP
  H
• INFO live.c(239) live_nfeatvec 13
• INFO main_live_pretend.c(92) PARTIAL HYP
  H E
• INFO live.c(239) live_nfeatvec 12
• INFO main_live_pretend.c(92) PARTIAL HYP
  H E
• INFO live.c(239) live_nfeatvec 13
• INFO main_live_pretend.c(92) PARTIAL HYP
  H E L
• INFO live.c(239) live_nfeatvec 12
• INFO main_live_pretend.c(92) PARTIAL HYP
  H E L
• INFO live.c(239) live_nfeatvec 13
• INFO main_live_pretend.c(92) PARTIAL HYP
  H E L OH

• Backtrace (null)
• LatID SFrm EFrm AScr LScr Type
• 254 0 45 -391470 -74100 -1
• 594 46 81 -472155 -148846 0 H
• 1291 82 102 -288621 -148846 0 E
• 1850 103 126 -235274 -148846 0 L
• 2599 127 147 -430694 -148846 0 L
• 2650 148 148 0 -148846 0
• 0 148 -1818214 -818330 (Total)
• FWDVIT H E L L (null)

6
Difficulties

• Hardware issues
• ASR software issues
• Letter phonemes
• Time

7
Solution
4 Stage Process
8
Solution

• Database (PostgreSQL)
• Names
• Phone numbers
• Fast access

9
Solution

• Architecture of application
• db.pm
• people.pm
• people.pl
• record.pl
• wav_to_raw.pl
• get_speech.pl
• display_speech.pm
• display_speech.pl
• VEPD.pm
• VEPD.pl

• Example
• PC press space bar before and after you speak
• User S AH EM
• PC Decoded as, SAM ?
• Results 1
• 1. SAM SMITH 765-973-2145

10
Solution
11
Results

• A first step towards hands free speech enabled
  phone directory
• Speaker Independent
• Applications Features
• Adding user
• Retrieving user (via speech)
• Manual search
• Viewing current phone directory

12
Possible Future Enhancement

• ASR enabled for
• Adding users
• Phone search
• Word Recognition (instead of letters)
• More accurate ASR (as tech. Grows)
• Graphical outlook (via perl/tk)
• Communication through VoiceXML

13
Special Thanks

• To friends and family
• Jim Rogers
• Hassan Halta
• Skylar Thompson
• Kushboo Goel
• Rabah family
• El-Shabab el-taybeh

14
Questions/Comments