app.py

from datasets import load_dataset
from transformers import pipeline
import soundfile as sf
import torch
import gradio as gr
import numpy as np

def predict_image(image):
  pipe = pipeline("image-classification", model="google/vit-base-patch16-224")
  ClassifedImage=pipe(image)

  result=ClassifedImage[0]['label']
  return result

def translate_to_arabic(text):
  pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-en-ar")
  result=pipe(text , max_length=100)
  return result[0]['translation_text']

def text_to_speech(text):
  pipe = pipeline("text-to-speech", model="MBZUAI/speecht5_tts_clartts_ar")
  embedding_dataset=load_dataset("herwoww/arabic_xvector_embeddings" , split="validation")
  speaker_embedding=torch.tensor(embedding_dataset[100]['speaker_embeddings']).unsqueeze(0)
  speech=pipe(text , forward_params={'speaker_embeddings':speaker_embedding})

  return (speech['sampling_rate'],np.array(speech['audio'], dtype=np.float32))

from PIL import Image
with gr.Blocks() as app:
  gr.Markdown("Image Classification, Arabic Translation, TTS")

  with gr.Row():
    with gr.Column():
      image_input=gr.Image(type="pil",label="Upload the Image to classify it" )
      classify_image=gr.Button("Classify the Image")
      pred=gr.Textbox(label="Classifcation Result")

    classify_image.click(fn=predict_image , inputs=image_input , outputs=pred)

  with gr.Row():
    translated_output=gr.Textbox(label="Translated Text")
    translate_btn=gr.Button("Translate to Arabic")

    translate_btn.click(fn=translate_to_arabic , inputs=pred , outputs=translated_output)

  with gr.Row():
    tts_btn=gr.Button("Convert to Speech")
    audio_output=gr.Audio(label="Audio Output")

    tts_btn.click(fn=text_to_speech , inputs=translated_output , outputs=audio_output)

app.launch()