import streamlit as st
from PIL import Image
import torch
from transformers import AutoImageProcessor
import pandas as pd
from transformers import ViTForImageClassification
from transformers import VitsModel, AutoTokenizer
import torch
from IPython.display import Audio

# Streamlit application title
st.title("Speech the Traffic Signs")

#Traffic Sign Classification
model= ViTForImageClassification.from_pretrained('Rae1230/Traffic_Signs_Classification')
processor = AutoImageProcessor.from_pretrained("google/vit-base-patch16-224")

uploaded_file = st.file_uploader("Choose a PNG image...", type="png", accept_multiple_files=False)
if uploaded_file is not None:
    img = Image.open(uploaded_file)
    st.image(img, caption='Uploaded Image.', use_column_width=True)
    inputs = processor(img.convert('RGB'), return_tensors="pt")

    outputs = model(**inputs)

    logits = outputs.logits
    img_class_idx=logits.argmax(-1).item()
    with open("labels.csv", "r") as file:
        df = pd.read_csv(file)
    
    
    num_col = df['ClassId']
    text_col = df['Name']
    
    text_value = text_col.loc[num_col == img_class_idx].values[0]
    st.write("Predicted class:", text_value)


#speech the Traffic Sign

    model2 = VitsModel.from_pretrained("facebook/mms-tts-eng")
    tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-eng")
    
    text = text_value
    inputs = tokenizer(text, return_tensors="pt")
    
    with torch.no_grad():
        output = model2(**inputs).waveform
    
    st.audio(output.numpy(),sample_rate=model2.config.sampling_rate)