Odessa is a basic speech recognition system which can identify very specific phrases of my own speech and respond accordingly, with......spunk.

This algorithmic detection of speech and signal processing were all done from scratch in Python through the use of numpy, scipy, matplotlib, sounddevice and soundfile libraries.

• All audio training samples for the phrases can be found here.
• All generated hmm binaries for the audio samples are found here.
• The generation of speech features from a sound signal is done via the asr_feature_builder.py.
• The training of hmm binaries from audio samples is done via the em.py.
• The execution of the program funneling identified live speech through recognition is done via the speech_recognizer.py.
• The detection of live speech segments and sampling to disk is done via the speech_sampler.py.
• The hot word tracking of odessa and subsequent phrase recognitions are done via the speech_state_machine.py.
• The funneling of audio samples in for training with different HMM parameters looking for an optimal configuration and generating "Training Results.xlsx" are done through trainer.py.

The final application looks like this!