Exploring the Web Speech API

Re:Exploring the Web Speech API by z3t4 on Saturday February 15, 2020 @06:19AM (#22334604)

SpeechRecognition worked great for me on Chrome for years until Google probably banned me for sending so many request. It would be cool with a completely offline SpeechRecognition... The trick with SpeechRecognition is to use grammars; if there are only a few words to choose from, it will more likely get it correct vs a random sentence.

[ Parent | on HN ]

Re:Exploring the Web Speech API by est31 on Saturday February 15, 2020 @07:38AM (#22334915)

> It would be cool with a completely offline SpeechRecognition

https://github.com/mozilla/deepspeech

This runs completely offline and is fast enough to run even on CPUs or smaller devices. Note though that it's not as accurate as an engine that requires TPUs for inferrence in order to be real time, which Google's engine probably does. World is full of tradeoffs...

[ Parent | on HN ]

Re:Exploring the Web Speech API by synesthesiam on Saturday February 15, 2020 @07:48AM (#22334979)

https://rhasspy.readthedocs.io/en/latest/

Rhasspy is fully offline and uses grammars to increase accuracy just like you said.

[ Parent | on HN ]

Re:Exploring the Web Speech API by interrealmedium on Saturday February 15, 2020 @06:47AM (#22334698)

>It would be cool with a completely offline SpeechRecognition...

Nuance still makes Dragon and Windscribe branded speech recognition software. Dragon was great +20 years ago. Not sure how it is now, but investigating offline speech recognition is on my agenda because Microsoft nerved theirs to push people to their online service.

[ Parent | on HN ]

Re:Exploring the Web Speech API by techbio on Saturday February 15, 2020 @09:19AM (#22335437)

Oh, gotcha, “Microsoft nerfed theirs”

[ Parent | on HN ]

Re:Exploring the Web Speech API by interrealmedium on Saturday February 15, 2020 @09:23AM (#22335458)

It worked fine in Windows 7, and barely gets "Yes" or "No" in its current iteration.

[ Parent | on HN ]