DeepSpeech as a (Docker) Service for IBus

This repository includes the following:

• a Docker container that can run DeepSpeech hardware-accelerated by your host OS's (nVidia) graphics card; the container reads audio from a pipe and reports results to an event
• an IBus Engine that allows the reslts of the recognition to be treated as regular input to the (Linux) host operating system

Installation/Setup

git clone https://github.com/mcfletch/deepspeech-docker.git
cd deepspeech-docker

Starting the daemon in a docker container:

./recogpipe/dockersetup.py
# when docker container is started...
./recogpipe/daemon.py

Feeding some data into the daemon from your microphone:

ffmpeg-audio.sh 

Feeding some data into the daemon from a wav file:

ffmpeg-sample.sh path/to/your/file.wav

Running the IBus daemon on your desktop:

./recogpipe/ibusengine.py

Testing bits out...

Sending audio to the daemon

./ffmpeg-audio.sh hw:1,1
./ffmpeg-sample.sh /path/to/some.wav

Viewing the raw transcripts from the daemon

nc -U /tmp/dspipe/events

Running the ibus engine interactively:

./ibusengine.py