audio_processing.py

import speech_recognition as sr  # For audio processing and recording audio
import wave  # To create a .wav file out of the microphone recordings
import whisper  # Advanced OpenAI audio transcribing model
import librosa  # Converting the .wav file into a floating-point time series
import numpy as np
import re


def nlp():
    recognizer = sr.Recognizer()

    try:
        with sr.Microphone() as source:
            audio_data = recognizer.listen(source, timeout=10)

        with wave.open("captured_audio1.wav", "wb") as wave_file:
            wave_file.setnchannels(1)
            wave_file.setsampwidth(audio_data.sample_width)
            wave_file.setframerate(audio_data.sample_rate)
            wave_file.writeframes(audio_data.frame_data)

        y, source = librosa.load("captured_audio1.wav")
        audio_waveform = np.array(y)

        audio_model = whisper.load_model("base")
        result = audio_model.transcribe(audio_waveform, fp16=False)["text"][1:]

        sen_type = None
        if result[-1] == "?":
            sen_type = "Interrogative"

        punctuation_pattern = re.compile(r'[^\w\s]')
        text = re.sub(punctuation_pattern, '', result)

        return sen_type, text

    except sr.UnknownValueError:
        print("Speech Recognition could not understand audio.")
        return None, None
    except sr.RequestError as e:
        print(f"Could not request results from OpenAI Whisper API service; {e}")
        return None, None