tdt_sc_8k

[creative64]

Hi,

Pocketsphinx provides a chinese acoustic model "tdt_sc_8k" in the
distribution. Want to get some basic info regarding the
model.
a) Is this a mandarin model or a cantonese one ?
b) There are 2 chinese dictionaries that come with distribution. Which one is
the dictionary compatible to this model ?
c) What is the encoding used in the dictionary (When I try opening it with MS
Words I see options like Unicode,
Unicode (Big endian), Unicode (UTF-7) and Unicode (UTF-8).
d) What kind of speech the model is trained on and its accuracy in general
(viz a viz hub4wsj_sc_8k english acoustic
model).
e) Where can I get the phoneme list and mdef.txt file for this model.

Thanks and regards,

[Nickolay V. Shmyrev]

a) Is this a mandarin model or a cantonese one ?

Mandarin

b) There are 2 chinese dictionaries that come with distribution. Which one
is the dictionary compatible to this model ?

Both are compatible, they are different by words orphography, not by the
sounds

c) What is the encoding used in the dictionary (When I try opening it with
MS Words I see options like Unicode,
Unicode (Big endian), Unicode (UTF-7) and Unicode (UTF-8).

UTF-8

d) What kind of speech the model is trained on and its accuracy in general
(viz a viz hub4wsj_sc_8k english acoustic
model).

It's trained on broadcast news. Accuracy is unknown and is not expected to be
high. It's more a model for small vocabularies.

e) Where can I get the phoneme list and mdef.txt file for this model.

You can convert binary mdef to text mdef with pocketsphinx_mdef_convert.

[creative64]

Thanks Nicole.

And now few more basic questions

a) For english phonemes the phonemeset is specified by
Phoneme Example Translation

---

AA odd AA D
AE at AE T
AH hut HH AH T
...etc

Is there a similar description available for phonemes used in chinese acoustic
model or it needs to be inferred by
looking at dictiionary ?

b) Is there an automated way to generate pronuciations for english words using
the phoneset present in this
chinese acoustic model (letter to sound conversion)?

Thanks and regards,

[Nickolay V. Shmyrev]

used in chinese acoustic model or it needs to be inferred by looking at
dictiionary ?

Is it a problem worth to ask?

Is there an automated way to generate pronuciations for english words using
the phoneset present in this chinese acoustic model (letter to sound
conversion)?

In Mandarin there is no regular alphabet so the pronunciation needs to be
predicted from a database. Unihan database could be used for that purpose. The
corresponding code (unihan_to_sphinx.pl) is a part of cmuclmtk

[creative64]

Thanks Nickole.

It would be great if in Windows build, pocketsphinx_mdef_convert is included
in projects so that
it compiles with other projects.

Thanks and regards,

[Nickolay V. Shmyrev]

It would be great if in Windows build, pocketsphinx_mdef_convert is included
in projects so that

Done in trunk

[creative64]

Thanks a lot Nickolay.

Regards,

PS: With the help of Google translate I'm able to create a small JSGF based
mandarin application
using tdt_sc_8k and it is working ok.

[Anonymous]

I was just looking for pocketsphinx_mdef_convert for windows as well, could
you point to where I can download it? I downloaded the latest snapshot from h
ttp://cmusphinx.sourceforge.net/wiki/download/, but it does not contain the executable.

Thank you.

[Nickolay V. Shmyrev]

I was just looking for pocketsphinx_mdef_convert for windows as well, could
you point to where I can download it? I downloaded the latest snapshot from h
ttp://cmusphinx.sourceforge.net/wiki/download/, but it does not contain the executable.

You need to build the tool yourself. You can use MS Visual Studio 2010 Express
for that.