A Java based voice assistant "test/prototype project" built with CMUSphinx and MaryTTS.

The objective was to develop a personal voice assistant, in order to improve upon project AVIA and AVII which were .NET Framework based, voice assistants that used SAPI. Their speech recognition accuracy lacked in all noisy environments. This project was an endeavour to explore other libraries and methods of speech recognition and TTS without the use of the .NET Framework while also having cross platform capabilities. Java proved to be an excellent choice as the compiled dependencies were included in the .jar file.

The project yielded interesting results, and at times performed better than SAPI, it was limited in adding voice commands and sentence recognition compared to SAPI. Thus, further development ceased since dealing with .gram files to create sentence structures was not desirable.

This project propelled me to learn neural networks as they would provide a solution where I needed to traverse a complex search space. This project was a beach head for project serena, which is now a private project under continued development, that uses multiple machine learning networks, that serve to power this new voice assistant.

For those of you still interested in JAVA based voice assistants, CMUSphinx and MaryTTS may prove useful to you.