Isolated word, speaker independent speech recognition Kaustubh R. Kale Guide: Dr. John G. Harris 12/7/2018 EEL 6825 project
Project Goals To make appliances smart Use Dynamic Time Warping algorithm 13 Mel frequency Cepstral coefficients as the extracted features Gui development and hardware interface 12/7/2018 EEL 6825 project
Description Schematic Diagram Endpoint detection in Java DTW analysis in Matlab Parallel Port operations via C++ Demo FOR MORE INFO... http://www.dcs.shef.ac.uk/~stu/com326/ 12/7/2018 EEL 6825 project
Schematic Diagram Java Matlab C++ Appliance 12/7/2018 EEL 6825 project
Endpoint detection in Java. Utterances are of unequal lengths Preceded by silence Use of signal power p[i..j] = k=i..j s[k]2 12/7/2018 EEL 6825 project
DTW analysis in Matlab Two basic concepts to be understood: 1. Feature extraction from the time dependant signal 2. Distance calculation: a.Local distance between features b.Global distance between signals 12/7/2018 EEL 6825 project
DTW Flow To obtain a global distance, time alignment must be done D(I,j)=min[D(I-1,j-1),D(I-1,j),D(I,j-1)] +d(I,j) 12/7/2018 EEL 6825 project
C++ interface with the port The matlab passes on the a parameter to the C++ program The C++ program drives the respective pins on the parallel port The Parameters: 1 = lights and fan off 2 = lights on and fan off 3 = fan on and lights off 4 = lights on and fan on 12/7/2018 EEL 6825 project
Classification Errors For speaker dependent operation the classification errors were 20% For speaker independent operation the classification errors were 30%-40% 12/7/2018 EEL 6825 project
Demonstration End to end operation 12/7/2018 EEL 6825 project
Future work Making the DTW more robust to ambient noise Achieving speaker independent word recognition Efficient inter component communication 12/7/2018 EEL 6825 project
Conclusion Via this program the goal of having voice operated smart appliance was achieved The error rate was around 20% 12/7/2018 EEL 6825 project
Thanks! Question time… 12/7/2018 EEL 6825 project