Any good architecture for continous inference

[JRMeyer]

>>> hnipun
[February 12, 2018, 9:24am]

Hi,

Is there any good resources to learn to implement a continuous speech to
text system using. From microphone sound capturing, segmenting the
continuous sound signal and any preprocessing methods till feed the
DeepSpeech model.

Thanks in advance.



[Uni-directional model for online/incremental ASR in real-time
applications e.g. voice assistants

[This is an archived TTS discussion thread from discourse.mozilla.org/t/any-good-architecture-for-continous-inference]

[JRMeyer]

>>> lissyx
[February 12, 2018, 9:44am]

You can probably have a look at the Voice Fill WebExtension, it's not
relying on DeepSpeech, but some people have provided DeepSpeech
'servers' implementation on Python and NodeJS.

[Archived Post]

[JRMeyer]

>>> Elleo
[February 13, 2018, 1:20pm]

You might like to take a look at the GStreamer DeepSpeech plugin I
wrote, which does some basic segmentation (just based on silence
thresholds):

Michael Sheldon slash 's Stuff -- 30 Dec
17

### Speech Recognition -- Mozilla's DeepSpeech, GStreamer and IBus

Recently Mozilla released an open source implementation of Baidu's
DeepSpeech architecture, along with a pre-trained model using data
collected as part of their Common Voice project. In an at...

[Archived Post]

[JRMeyer]

>>> hnipun
[February 14, 2018, 4:18am]

Thanks a lot. Its a really nice idea. Do you have any documentation so
that I can learn about your architecture.

[Archived Post]