Voice recognition for performing detection and recognition
This is still considered a non-official release.
libsndfile fftw3
For debugging purposes, also install the google CPU profiler.
In the top-level directory, type make.
- feature_extractor - Includes a library that perform Mel-Frequency Cepstral Coefficient extraction from raw normalized speeech.
- recognizer - Classifiers for signal processing. Currently, the only classifier is Dynamic Time Warping.
- server - Setup a socket server which can receive JSON-formatted speech.
- bin - Symlinks to various
- reference - Location of stored features.
- scripts - Location of bash-scripts which bares the same name as the their assoiated features in referece/
$: ./bin/server 192.168.0.100 54321
When receiving, the server store the speech signal intermittently in a file in the top-level directory called sample.wav. After processing sample.wav the feature extractor stores a file in the top-level directory called features.json which may be saved to reference/ and given a proper name. Upon a successful lookup into reference, the classifier will try to run a bash script that has the same name in the scripts directory.
- Port is fixed in source to 54321
- A high-pass filter is recommended for optimal training.