Two types of examples are being demonstarted in DL based End to end speech recognition. Phoneme recognition and character level recognition,
Is there any advantage of phoneme recognition over character recognition
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Spelling to pronunciation rules can be very arcane in most languages,
and you end up using much of the capacity of your network to capture
these oddities. So in that sense, phoneme recognition is the more
natural task.
By the same token, though, the spelling oddities, if well captured,
can end up providing you with a stronger grammar than just phonetic
structure.
I expect someone has run the comparison, although I haven't seen any
myself. Perhaps I can have Vishal run this test; he's currently
obtaining ~10% CER on WSJ
Two types of examples are being demonstarted in DL based End to end speech
recognition. Phoneme recognition and character level recognition,
Is there any advantage of phoneme recognition over character recognition
Two types of examples are being demonstarted in DL based End to end speech recognition. Phoneme recognition and character level recognition,
Is there any advantage of phoneme recognition over character recognition
Its iffy.
Spelling to pronunciation rules can be very arcane in most languages,
and you end up using much of the capacity of your network to capture
these oddities. So in that sense, phoneme recognition is the more
natural task.
By the same token, though, the spelling oddities, if well captured,
can end up providing you with a stronger grammar than just phonetic
structure.
I expect someone has run the comparison, although I haven't seen any
myself. Perhaps I can have Vishal run this test; he's currently
obtaining ~10% CER on WSJ
-Bhiksha
On Tue, Jul 4, 2017 at 12:40 AM, Pankaj pankaj2701@users.sf.net wrote:
--
Bhiksha Raj
Carnegie Mellon University
Pittsburgh, PA, USA
Tel: 412 268 9826