perm filename NOTE[11,ALS] blob
sn#064423 filedate 1973-09-27 generic text, type T, neo UTF8
00100 Speech Recognition
00150
00200 There is an active project on Speech Recognition with the main
00300 emphasis having to do with, 1) the extraction of phonetically
00400 significant information from the acoustic signal and 2) the use of a
00500 machine learning scheme for correlating these acoustic data with the
00600 linguistic content of the utterance.
00700
00800 Possible student projects might be-
00900
01000 1. The development of a routine for the extraction of any one of a
01100 number of different significant features of the acoustic wave. This
01200 would involve a literature search relating to what has been done, the
01300 development of a proposed scheme to overcome the objections to the
01400 best that has been done, the writing of code to implement the new
01500 idea and the testing of this code on our existing corpus of speech
01600 data.
01700
01800 2. Assistance in the work of annotating our increasing corpus of
01900 speech data so that it can be used for machine learning runs. This
02000 would force the student to become conversant with the conventional
02100 phonetic symbols, with our particular variations to these conventions
02200 which have been made to relate them more directly to the actual
02300 acoustic signals and he would have to add to our library of routines
02400 for the easy manipulation of speech files and for the linking of the
02500 lists of phonetic symbols with the acoustic data.
02600
02700 3. The refinement and possible improvement of the present Signature
02800 Table approach to machine learning as applied to speech recognition,
02900 or the development of an entirely new method to replace the existing
03000 one. This would be a rather ambitious project but it could easily
03100 lead to a very good thesis topic.
03200
03300 4. The study of the effects of context on the acoustic representation
03400 of intended phonetic utterances and the development of methods to
03500 make use of these contextual modifications as an aid to understanding
03600 the phonetic intent.