A DeepSpeech webserver with resampling capabilities that run over websockets through Starlette.

• Python 3.6+
• Starlette 0.12
• Uvicorn 0.7
• Gunicorn 19.9
• Scipy 1.3
• Webrtcvad 2.0

I'm using miniconda so:

$ conda create --name wsDSpeech python=3.6 # or 3.7
$ conda activate wsDSpeech
$ pip install -r requirements.txt

First, set up the correct INPUT_SAMPLE_RATE in appConfig according to the audio file that will be streamed to the server.

Then, if you want to use the Dockerfile just run:

$ docker build -t deepspeech .
$ docker run -t deepspeech

or locally:

$ cp appConfig app/appConfig
$ gunicorn -c guConfig.py app:app

Now you can reach the server at:

http://127.0.0.1:5000

for a cutie hello world

Meanwhile a thread starts and downloads DeepSpeech models if not present.

In the example folder there are two simple clients.

• Choose how many connections:

   nrClient=2

• Just run it

   cd examples/python
   python simpleClient.py

• appConfig contains all the settings used.

• On first run, DeepSpeech models will be downloaded under the DEEPSPEECH_ROOT_PATH.

• If you want more verbosity, flag VERBOSE/DEBUG to True

• guConfig contains all the settings related to GUnicorn, including the deepspeech model downloading part.

• Python client example with more connections
• Manage memory leaking (avoid to force workers restart)
• Handle async/future exceptions
• Use a lighter docker as base image (Alpine?)