simudream / long-audio-aligner

http://sourceforge.net/p/cmusphinx/code/11633/tree/branches/long-audio-aligner/Aligner

CMUSphinx Long Audio Aligner

AUTHORS

• Bhiksha Raj : CMUSphinx Mentor GSoC 2011

• Nickolay Shymrev : CMUSphinx Mentor GSoC 2011

• James Baker : CMUSphinx Advisor GSoC 2011

• Antony Robinson : CMUSphinx Advisor GSoC 2011

• Rita : CMUSphinx Advisor GSoC 2011

• Apurv Tiwari : Student GSoC 2011

ABOUT

Speech and text alignment finds applications in fields such as multimedia indexing and training of large vocabulary speech recognition. This project is designed to provide this functionality in CMUSphinx's Sphinx4 decoder.

Result of a typical audio - text alignment is a sequence of word indexed with their start and end time in the audio.

REQUIREMENTS

i) JDK version >= 1.6.x.x ii) Sphinx4.jar

RUN INSTRUCTIONS

To run the demo :

i) Build the program by issuing

$ ant all

This will give you bin/aligner.jar

ii) Run it by doing

$ java -ms400m -mx1500m -jar bin/aligner.jar <path-to-audio-file> <path-to-transcript-file>

UNDERSTANDING THE RESULT OF ALIGNMENT

Result of alignment is displayed on standard output. Result will contain of a sequence of words, each followed by it's start and end time in the utterance as detected by Long Audio aligner.

Eg. Sphinx(1.01,1.37)

Word detected : Sphinx Start Time : 1.01 sec End Time : 1.37 sec