A Python wrapper for simple offline real-time dictation (speech-to-text) and speaker-recognition using Vosk. Check out the official Vosk GitHub page for the original API (documentation + support for other languages).

This module was created to make using a simple implementation of Vosk very quick and easy. It is intended for rapid prototyping and experimenting; not for production use.

For example, I used this module in a quick personal-assistant program.

• Uses Vosk: lightweight, multilingual, offline, and fast speech recognition.
• Runs in background thread (non-blocking).
• Both complete-sentence and real-time outputs.
• Optional speaker-recognition (using X-Vectors).
• Configurable filter-phrase list (eliminate common false outputs).

Should work with Python 3.6+. Tested with Python 3.8.7 on Windows 10 1903.

Python Modules: (see requirements.txt)

• vosk
• sounddevice
• numpy

You will also need to download Vosk models; one for your language of choice, and (if desired) the speaker-recognition model. Both can be found on the Vosk models page. If you don't use speaker recognition, you only need the one model.

This repository contains some examples of usage; ExampleSimpleDictation.py, ExampleSpeakerRecognition.py, and ExampleNonBlocking.py. Check the Documentation.md file for more in-depth info.

Below is the simplest implementation to get a fully-functioning speech-recognition system.

import simpleVosk as sv

def prnt(txt, spk, full):
	print(txt)

s = sv.Speech(callback=prnt, model="model")
s.run(blocking=True)

Make sure your default input device is working, and/or ensure you are passing the correct DeviceID to the Speech object. You can see device IDs with the listDevices() method in simpleVosk.py.

Make sure you have Windows microphone access enabled. Having this disabled can cause errors similar to this: sounddevice.PortAudioError: Error opening RawInputStream: Unanticipated host error [PaErrorCode -9999]: 'Undefined external error.' [MME error 1]

This project goes against some standard Python conventions:

• It uses camelCase for naming methods (and files) rather than snake_case
• Tabs are used rather than 4 spaces for indentation (as I am a sane human being)
• Non-standard docstring formats are being used

• Add ability to add custom words/phrases (KaldiRecognizer appears to only accept replacement dictionaries)
• Use proper docstrings