Installing the free speech model

If your application uses free speech recognition, in other words, speech recognition without the need to create a grammar, you need to install and configure the package with the free speech model. Follow the instructions:

1) Run the model installer

Run the file slm-VERSION.run. See the example:

$ sudo ./slm-ptbr-1.0.1-am-ptbr-8k-1.0.6-rc.run

Verifying archive integrity...  100%   All good.
Uncompressing slm-ptbr-1.0.1  100%

Installing statistical language model (SLM) [slm-ptbr-1.0.1]
Success!

2) Configure the model to be loaded

Configure the CPQD ASR to load this model at system startup. Edit the file /opt/cpqd/asr/config/engine/engine.conf and change the key --lm.preloadModels, to include the list of models, separated by commas. When we install the general free speech package slm-ptbr-1.0.1-am-ptbr-8k-1.0.6-rc.run, its URI is builtin:slm/general, thus, its configuration must be as shown in the following example:

# ==================================
# Configure recognition
# ==================================

--lm.preloadModels=builtin:slm/general

#--decoder.maxSentences=1
#--decoder.confidenceThreshold=30

Warning

The key --lm.preloadModels is generally configured as shown above, but comes with the comment “#”. In this case, simply remove the “#” symbol.