emp7eror / deep-talk Goto Github PK

Deep-speech react app to test trained models,to visualize the speech to text process, to record the audio from mic to wav using the webaudio API, or to create/use a custom open speech-to-text API.

Live Demo.

git clone https://github.com/buddyeorl/deep-talk.git

This app needs two files to work, the acoustic model: deepspeech-0.8.2-models.pbmm

and the following scorer: deepspeech-0.8.2-models.scorer

download both files to /server/index/

In the terminal in the repo root directory npm run build .

Now in the terminal cd server && node server.js

*** defaults to http://localhost:3001 ***

API calls to

https://deep-talk.azurewebsites.net/api/v1/getVoice

POST requests accepts 16kHZ mono 16bits WAV audio files in multipart form data,the field name should be 'audio'

sample responses:

No audio file:

{
    "message": "No audio file has been received"
}

No recognition:

{
    "error": "No speech was recognized"
}

Success:

{
    "message": "success",
    "data": "two three"
}

Please note that the app resamples the audio recorded to 16kHZ mono 16bits(as used when training the model), I might add different samplerates recording options if requested.

Also note that this app will recognize pauses in speech and trimm the audio files and speech recognition responses accordingly.

Github Alex Lizarraga

Portfolio www.alexcode.io

Email [email protected]