4 Thanks: David Huggins-Daines, Chen Tao, creative64, Edwin Miguel, Bhiksha Raj
8 * Nice performance improvements
9 * Use arc iterators to build FSG lextrees, speeds this up immensely
10 * Filler word propagation is responsible for 5% speedup
11 * Continuous decoder can decode files, not from the microphone
12 * Add senone score dumping and reloading capability
13 * Functional alignment pass
15 * Add the Android makefile and other JNI stuff
16 * Case sensitive names, including phone names
17 * Better error messages
18 * Report timing for individual search passes
22 * Accuracy regression fixes
23 * Fixes the bug with reading transition matrix on big-endian platforms
24 * Fixes very important regression with NULL transitions in fsg search
25 * Proper acoustic score scaling during posterior calculation.
27 And many, many, many more intersting things!
32 2000-01-27 CMU Sphinx is going Open Source!