After this server issue crippled Google’s Home assistant (and some people could not even turn on the lights), I realized the value of OFFLINE speech recognition. And now we see in the news that recordings are being permanently stored and reviewed by humans for training purposes at all of the cloud-based speech recognition providers. So I would like to bring these offline alternatives to your attention:

– https://pypi.org/project/SpeechRecognition
– https://www.quora.com/What-are-the-top-ten-speech-recognition-APIs
_____

"Mozilla researchers aim to create a competitive offline STT engine called Pipsqueak that promotes security and privacy. This implementation of a deep learning STT engine can be run on a machine as small as a Raspberry Pi 3."

https://research.mozilla.org/machine-learning
https://blog.mozilla.org/blog/2017/11/29/announcing-the-initial-release-of-mozillas-open-source-speech-recognition-model-and-voice-dataset
https://voice.mozilla.org/en

Mozilla DeepSpeech is a TensorFlow implementation of Baidu’s DeepSpeech architecture
https://github.com/mozilla/DeepSpeech

"Our word error rate on LibriSpeech’s test-clean set is 6.5%, which gets us close to human level performance."
— "Yes, this model can be used to do offline speech recognition."
https://hacks.mozilla.org/2017/11/a-journey-to-10-word-error-rate

comment: if the error rate of IBM’s online engine is 5.5% , you can see how impressive Baidu’s engine is, to achieve a 6.5% error rate on a Raspberry Pi (!)
___________

PS: at first glance I thought the Blinkt device was kind of a gimmick — but now I think I might want that, if I could make each lamp to be an activity indicator for a different extension (plus a couple lights for channel activity on a 2-channel SIP trunk.) This could be helpful for troubleshooting purposes, and would also remind me not to commit any changes in the PBX user interface until all extension & channel activity has ceased. 😉