The pocketsphinx-sonic-server is a wrapper over the PocketSphinx speech recognition system which allows us to communicate with it using the Sonic protocol. PocketSphinx itself is a small-footprint continuous speech recognition system, suitable for handheld and desktop applications. We can use a client such as AcquireSpeech in order to communicate with pocketsphinx-sonic-server and get the text for the speech.
It is fairly easy to set up a language model for PocketSphinx and use it for speech recognition.
- Location: /core/pocketsphinx-sonic-server/
- Language: C++
- Distribution: Binary
- Platform(s): Windows
Using command line parameters
The applications expects a config file to be specified on the command line as follows
- -c [file-name]
This config file should contain the following information specified as below
- -lm [the language model file to be used]
- -dict [the dictionary to be used]
- -hmm [the acoustic mode]
By default, the Virtual Human Toolkit uses the wall street journal acoustic model that comes with pocketsphinx and the CMU pronunciation dictionary. You can change this to use your own.
- -samprate [the sampling rate]
Creating a language model for PocketSphinx Wrapper
You will need to follow the below steps for creating your own language model for use with the PocketSphinx Wrapper.
- Go to the "data\pocketsphinx" folder within the Toolkit
- There is a file called corpus.txt which contains the utterances/lines to be spoken by the character. Replace this file with a file containing the new lines which you want the character to speak. Each utterance is one line in the file
- Now double click the batch file called generate_language_model.bat
- This will generate a new lm.arpa file which will be the new language model containing the lines in corpus.txt
- Now if you relaunch the PocketSphinx Wrapper from the launcher, it will reference this new language model that you just created
See Main FAQ for frequently asked questions regarding the installer. Please use the Google Groups emailing list for unlisted questions.