In the first article, we talk and building a speech recognition system it uses the internet to connect to google and use its speech recognition algorithm, today in this article we going to build a speech recognition system when you are offline.

let's get started.

First of all, there is a python library called, VOSK. to install it on your computer type this command

pip3 install vosk

for more details please visit: https://alphacephei.com/vosk/install

now we need to install pyaudio:

pip install pyaudio

now our pre-requirements are done, now we have to build the offline speech recognition program. first, we need to import all the necessary

from vosk import Model, KaldiRecognizer
import pyaudio

here what we do is import the model and the kaldiRecognizer from the vosk library. along with that, we import py audio.

now we have to download the model for that go to this website and choose your preferred model and download it: https://alphacephei.com/vosk/models

here I use vosk-model-small-en-us-0.15 as my model

after downloading, you can see it is a compressed file unzip it in your root folder, like this

speech-recognition/
    ├─ vosk-model-small-en-us-0.15 ( Unzip follder ) 
    ├─ offline-speech-recognition.py ( python file )

now create a variable called " model " and type this

model = Model(r"C:\\Users\User\Desktop\python practice\ai\vosk-model-small-en-us-0.15")

what we do here we create a variable called " model " and assign our unzip folder location with the Model

speech-recognition/
    ├─ vosk-model-small-en-us-0.15 ( Unzip follder )

now we going to create a variable called " recognizer " and assign KaldiRecognizer

recognizer = KaldiRecognizer(model, 16000)

now as usual we have to import the microphone to our code so we import pyaudio

now we going to call the pyaudio to use the microphone to catch the voice.

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

now we enter the fun part :

while True:
    data = stream.read(4096)


    if recognizer.AcceptWaveform(data):
        text = recognizer.Result()
        print(f"' {text[14:-3]} '")

what we do here is write a while loop

create a variable called data and assign it to a " stream.read(4096) "

    data = stream.read(4096)

now we create a decision " if " recognizer caught a waveform in the data

then assign the results to the variable called " text " and print it out.

    if recognizer.AcceptWaveform(data):
            text = recognizer.Result()
            print(text)

this is the output we got :

    {
      "text" : "hello"
    }
    {
      "text" : "hello"
    }
    {
      "text" : "hi"
    }

to remove { and "text" changes the print command to this :

    print(f"' {text[14:-3]} '")

now the output will look like this, clean and more readable:

    '  '
    ' hello '
    ' hi '
    '  '

the complete code :

    from vosk import Model, KaldiRecognizer
    import pyaudio

    model = Model(r"C:\\Users\User\Desktop\python practice\ai\vosk-model-small-en-us-0.15")
    recognizer = KaldiRecognizer(model, 16000)

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

    while True:
        data = stream.read(4096)


        if recognizer.AcceptWaveform(data):
            text = recognizer.Result()
            print(f"' {text[14:-3]} '")

visit https://alphacephei.com/vosk/models for more languages and models.