Dynamic Grammar/Language Model

[Anonymous]

Hello,

I need to implement a simple command-and-control voice recognition setup that
includes being able to play multimedia fies, e.g. "Play Sweet Child of Mine".
Is there a way to update an existing grammar or language model at runtime? The
existing multimedia files may change at any time and the application needs to
know which song names are valid.

Thanks,
Shea

[Shredder Woods]

Hello Shea,
Yes you can update the existing grammar or language model at runtime.
Suggested reading: Sphinx 4 decoder model. topic "GRAPH CONSTRUCTION MODULE".

The graph construction module in Sphinx-4, i.e. the linguist, has two
submodules.
The first interprets the language model provided by the application as a part
of the
KB, and converts it into an internal grammar. The internal grammar
representation
of word lists links a single source node to all words, and has edges from the
outputs
of words to a common sink node. A loopback is made from the sink to the
source.
For N-gram LMs, a fully-connected structure is formed where every word is
represented by a node, and there are explicit links from every node to every
other
node.
The internal grammar is then converted to a language HMM by the second
submodule, which is independent of the grammar construction module. In forming
the language HMM, the word-level network of the internal grammar is expanded
using the dictionary and the structural information from the acoustic models.
The
acoustic models for the sub-word units are incorporated into the final
language
HMM.

Also have a look at my thread.
https://sourceforge.net/projects/cmusphinx/forums/forum/5471/topic/4381140

Regards,
Shredder