Is it possible to add phones to an existing Sphinx acoustic model? I know in HTK it's possible to copy a phone model and allow the copy to diverge from the original through training. Is there a tool that will duplicate a phone like that for Sphinx models? I'm wanting to use an existing background model such as WSJ1 or Hub4, and adapt it using an Australian corpus using a modified version of the BEEP dictionary. Unfortunately BEEP uses a larger phone set than the pre-built Sphinx models: there are 5 extra phones (ea, ia, oh, ua, ax). I'm not a linguist of any sort, but it seems short "o", which is common is Australian English, has disappeared from US English, replaced by long "a" - or so pronunciation dictionaries would have us believe. In any case, are there any such acoustic model hacking tools for Sphinx? - thanx.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Is it possible to add phones to an existing Sphinx acoustic model? I know in HTK it's possible to copy a phone model and allow the copy to diverge from the original through training. Is there a tool that will duplicate a phone like that for Sphinx models? I'm wanting to use an existing background model such as WSJ1 or Hub4, and adapt it using an Australian corpus using a modified version of the BEEP dictionary. Unfortunately BEEP uses a larger phone set than the pre-built Sphinx models: there are 5 extra phones (ea, ia, oh, ua, ax). I'm not a linguist of any sort, but it seems short "o", which is common is Australian English, has disappeared from US English, replaced by long "a" - or so pronunciation dictionaries would have us believe. In any case, are there any such acoustic model hacking tools for Sphinx? - thanx.