VOSK the Offline Speech Recognition

Introduction

What is VOSK? VOSK is a powerful tool for real-time speech recognition that does not require an internet connection. Developed by Alpha Cephei, it supports multiple languages and is highly efficient. It can even run on low-performance devices, such as the Raspberry Pi.

Using VOSK

To use VOSK, you first need to download the model from this website: https://alphacephei.com/vosk/models

VOSK offers models for many languages, including Portuguese, English, Japanese, and others:

Once you download the ZIP file containing the VOSK model, you will need to unzip it

---

Install Dependences

First, you need to install the VOSK package and PyAudio using the following command:

pip install vosk pyaudio

PyAudio is a library that allows you to capture and reproduce audio using the PortAudio API. In this code, PyAudio is used to:

• Open the microphone.
• Capture audio in real time.
• Collect audio data to Vosk for processing.

---

Let's Code!!!

Code Exemple:

import pyaudio
import json
from vosk import Model, KaldiRecognizer

model = Model("vosk-model-en-us-0.22")
recognizer = KaldiRecognizer(model, 16000)

p = pyaudio.PyAudio()
stream = p.open(format=pyaudio.paInt16, channels=1, rate=16000, input=True, frames_per_buffer=8192)
stream.start_stream()

print("listening...")

while True:
    data = stream.read(4096, exception_on_overflow=False)
    if recognizer.AcceptWaveform(data):
        result = json.loads(recognizer.Result())
        print("You:", result["text"])

I created this simple example code to demonstrate how VOSK works.

1. Library Imports

import pyaudio
import json
from vosk import Model, KaldiRecognizer

It's a simple import library

2. Speech recognition model loading

model = Model("vosk-model-en-us-0.22")

This loads the VOSK model for English (US).
3. Speech recognizer initialization

recognizer = KaldiRecognizer(model, 16000)

• KaldiRecognizer: Starts Speech Recognition model with 16000 Hz (16 kHz) sampling rate, commum sampling rate for Speech Recognition

4. Audio Settings

p = pyaudio.PyAudio()
stream = p.open(format=pyaudio.paInt16, channels=1, rate=16000, input=True, frames_per_buffer=8192)
stream.start_stream()

• pyaudio.PyAudio(): Initializes the PyAudio object, which is used to configure and manage audio capture.

• p.open(): Opens an audio stream with the following settings:

• format=pyaudio.paInt16: Audio format of 16 bits.

• channels=1: Mono audio (1 channel).

• rate=16000: Sampling rate of 16000 Hz.

• input=True: Indicates that the stream will be used for audio input (capture).

• frames_per_buffer=8192: Size of the audio buffer.

• stream.start_stream(): Starts the audio capture.

5. Speech Recognition Loop

print("listening...")

while True:
    data = stream.read(4096, exception_on_overflow=False)
    if recognizer.AcceptWaveform(data):
        result = json.loads(recognizer.Result())
        print("You:", result["text"])

• stream.read(4096, exception_on_overflow=False):
  Reads 4096 frames of audio from the stream. The exception_on_overflow=False parameter prevents the program from raising an exception in case of overflow.

• recognizer.AcceptWaveform(data): Sends the audio data to the speech recognizer. If the recognizer detects a complete phrase, it returns True.

• json.loads(recognizer.Result()): Converts the speech recognition result (which is in JSON format) into a Python dictionary.

Conclusion

VOSK is a powerful and efficient tool for real-time speech recognition, supporting multiple languages and running seamlessly on low-performance devices like the Raspberry Pi.
Its offline capability makes it ideal for applications where internet access is limited.
With this guide, you’ve learned how to set up VOSK, configure audio capture, and implement a basic speech recognition system.
Whether for voice-controlled apps, transcription tools, or language learning, VOSK offers a simple yet robust solution for integrating speech recognition into your projects.

---

Thanks for reading 🙃