app.py

import streamlit as st
from audiorecorder import audiorecorder
from transformers import AutoModelForAudioClassification, AutoFeatureExtractor
import torch
import librosa
from scipy.io.wavfile import read as read_wav
from io import BytesIO
import soundfile as sf
import io
from PIL import Image
import pandas as pd
import torch.nn as nn

import yaml
from yaml.loader import SafeLoader


import streamlit as st
import streamlit_authenticator as stauth
from streamlit_authenticator import Authenticate

hugging_face_model = "MeshalAlamr/wav2vec2-xls-r-300m-arabic_speech_commands"

@st.cache(allow_output_mutation=True)
def load_model():
    feature_extractor = AutoFeatureExtractor.from_pretrained(hugging_face_model)
    model = AutoModelForAudioClassification.from_pretrained(hugging_face_model)
    return model, feature_extractor

model, feature_extractor = load_model()

# st.write(hashed_passwords)

with open('config.yaml') as file:
    config = yaml.load(file, Loader=SafeLoader)

authenticator = Authenticate(
    config['credentials'],
    config['cookie']['name'],
    config['cookie']['key'],
    config['cookie']['expiry_days'],
    config['preauthorized']
)

name, authentication_status, username = authenticator.login('تسجيل الدخول', 'main')


if st.session_state["authentication_status"]:
    st.write(f'مرحبا، *{st.session_state["name"]}*')
    

    st.title("التعرف على الأوامر العربية")
    
    @st.cache(allow_output_mutation=True)
    def load_model():
        feature_extractor = AutoFeatureExtractor.from_pretrained(hugging_face_model)
        model = AutoModelForAudioClassification.from_pretrained(hugging_face_model)
        return model, feature_extractor

    model, feature_extractor = load_model()

    audio = audiorecorder("اضغط هنا للتسجيل", "يتم التسجيل... اضغط لإيقاف التسجيل")


    english_to_arabic = {
    'backward' : 'خلف',
    'cancel' : 'إلغاء',
    'close' : 'إغلاق',
    'digit' : 'رقم',
    'direction' : 'اتجاه',
    'disable' : 'تعطيل',
    'down' : 'أسفل',
    'eight' : 'ثمانية',
    'enable' : 'تفعيل',
    'enter' : 'إدخال',
    'five' : 'خمسة',
    'forward' : 'أمام',
    'four' : 'أربعة',
    'left' : 'يسار',
    'move' : 'تحريك',
    'next' : 'التالي',
    'nine' : 'تسعة',
    'no' : 'لا',
    'ok' : 'موافق',
    'one' : 'واحد',
    'open' : 'فتح',
    'options' : 'خيارات',
    'previous' : 'السابق',
    'receive' : 'استقبال',
    'record' : 'تسجيل',
    'right' : 'يمين',
    'rotate' : 'تدوير',
    'send' : 'إرسال',
    'seven' : 'سبعة',
    'six' : 'ستة',
    'start' : 'ابدأ',
    'stop' : 'توقف',
    'three' : 'ثلاثة',
    'two' : 'اثنان',
    'undo' : 'تراجع',
    'up' : 'أعلى',
    'yes' : 'نعم',
    'zero' : 'صفر',
    'zoom in' : 'تكبير',
    'zoom out' : 'تصغير',
    }


    if len(audio) > 0:
        # To play audio in frontend:
        st.audio(audio)
        # To save audio to a file:
        wav_file = open("temp_audio.wav", "wb")
        wav_file.write(audio.tobytes())

        classify = st.button("اضغط هنا للتعرف")
        if classify:

            array, sampling_rate= librosa.load("temp_audio.wav", sr=48000)


            array = librosa.resample(array, orig_sr = sampling_rate, target_sr = 16000)

            input_audio = feature_extractor(array,
                                sampling_rate=16000, padding=True, return_tensors="pt")
            
            softmax = nn.Softmax(-1)       
            logit = model(input_audio['input_values']).logits
            predicted_id = int(torch.argmax(logit, dim=-1))
            confidence_score = str(round(torch.max(softmax(logit)).item()*100,2))
            st.subheader("الكلمة المتوقعة" + ": "  + english_to_arabic[model.config.id2label[predicted_id]] )
            st.subheader("%" + "نسبة التأكد" + ": "  + confidence_score)

    classes = [english_to_arabic[i] for i in [v for v in model.config.id2label.values()]]

    if len(classes)%2 == 0:
        df = pd.DataFrame({"الأمر" : classes[:len(classes)//2],
                    " الأمر" : classes[len(classes)//2:]})
    else:
        df = pd.DataFrame({"الأمر" : classes})
        
    # CSS to inject contained in a string
    hide_table_row_index = """
                <style>
                thead tr th:first-child {display:none}
                tbody th {display:none}
                </style>
                """

    # Inject CSS with Markdown
    st.markdown(hide_table_row_index, unsafe_allow_html=True)

    # Display a static table
    st.header("الأوامر المتوفرة")
    st.table(df)

    authenticator.logout('تسجيل الخروج', 'main')



elif st.session_state["authentication_status"] == False:
    st.error('اسم المستخدم أو كلمة المرور خاطئة')
elif st.session_state["authentication_status"] == None:
    st.warning('أدخل اسم المستخدم وكلمة المرور')