app.py

import pandas as pd
import difflib
import gradio as gr
from transformers import pipeline
import librosa
import re

# import numpy as np

transcriber = pipeline("automatic-speech-recognition", model="openai/whisper-large-v3")


class Model_Voice_Text():
    
    """
    This class takes the voices, convert them to text
    """
    #open and read the file after the appending:

    def __init__(self) -> None:
        # self.SR_obj = sr.Recognizer()
        self.KEYWORDS = ['suicide', 'urgent', 'poor', 'in-need', 'old', 'pregnant', 'refugee', 'new immigrant', 'patient', 'ill', 'sick', 'anxiety', 'anxious']
        self.sins = [5678, 1967, 4530, 3986, 9750, 1065, 7134, 6410, 2906, 8056, 1307, 3503, 7708, 4980, 1248, 3491, 6157, 9242, 3198, 5632]
        # self.fuzzer = fuzz.Fuzz()
    
    # Define a function to find the number of times the word similar to the word stored in variable target_var, in a text stored in a variable named text_res
    def find_similar_word_count(self, text, target_var):
        """Finds the number of times the word similar to the word stored in variable target_var, in a text stored in a variable named text_res using difflib.

        Args:
            text: The text to search.
            target_var: The word to find the similar word to.

        Returns:
            The number of times the word similar to target_var appears in the text.
        """

        # Create a list of all words in the text
        words = text.split()

        # Find all words similar to target_var
        similar_words = difflib.get_close_matches(target_var, words, cutoff=0.75)

        # Return the number of similar words
        return len(similar_words)
    
    def extract_phone_number(self, text):
        # Define a regular expression pattern to match phone numbers
        phone_pattern = re.compile(r'\b\d{7,}\b')

        # Search for the phone number in the text
        match = re.search(phone_pattern, text)

        # Check if a match is found and return the phone number
        if match:
            return match.group()
        else:
            return "000"
        
    def extract_sin(self, text):
        # Define a regular expression pattern to match phone numbers
        sin_pattern = re.compile(r'\b\d{4}\b')

        # Search for the phone number in the text
        matches = re.findall(sin_pattern, text)
        if matches:
            return matches 
        else: return "Not detected"

    def check_eligibility(self, sins_ex):
        for number in sins_ex:
            if number in self.sins:
                return "Eligible"
        return "Not Eligible"

    def matching_text(self, text):
        df = pd.DataFrame()
        ph_num = '000'
        sin = '0000'
        ret = []
        # words = nltk.word_tokenize(text)
        for target_var in self.KEYWORDS:
            count = self.find_similar_word_count(text, target_var)
            
            # matches = process.extract(text, word)
            if count>0:
                ret.append(target_var)
                ret.append(count)
        if ret == []:
            ret.append("nothing found")
        
        ph_num = self.extract_phone_number(text=text)

        sin = self.extract_sin(text=text)

        eligib = self.check_eligibility(sins_ex=sin)
        
        # initialize data of lists. 
        data = {'Keywords': [ret], 
                'Phone Number': ph_num,
                'SIN': sin,
                'Eligible': eligib,
                'text': text} 
        df = pd.DataFrame(data)
        
        # ret.append(text)
        return df
    
    def transcribe(self, audio_f):
        text = ""

        # First load the file
        audio, sr = librosa.load(audio_f)

        # Get number of samples for 20 seconds; replace 20 by any number
        buffer = 20 * sr

        samples_total = len(audio)
        samples_wrote = 0
        counter = 1

        while samples_wrote < samples_total:

            #check if the buffer is not exceeding total samples 
            if buffer > (samples_total - samples_wrote):
                buffer = samples_total - samples_wrote

            block = audio[samples_wrote : (samples_wrote + buffer)]
            text += transcriber(block)["text"]
            counter += 1
            samples_wrote += buffer

        return text
    
    def voice_to_text_s(self, audio):
        tran_text = self.transcribe(audio)
        # print(tran_text)
        match_results = self.matching_text(tran_text.lower())
        return match_results


model = Model_Voice_Text()


demo = gr.Blocks()


micro_ph = gr.Interface(fn=model.voice_to_text_s,
             inputs=gr.Audio(source="microphone", type="filepath"),
             outputs=gr.Dataframe(label="Output Box", interactive=True))

file_ph = gr.Interface(fn=model.voice_to_text_s,
             inputs=gr.Audio(source="upload", type="filepath"),
             outputs=gr.Dataframe(label="Output Box", interactive=True))


with demo:
    gr.TabbedInterface(
        [micro_ph, file_ph],
        ["Transcribe Microphone", "Transcribe Audio File"],
    )

demo.launch(debug=True)