from transformers import pipeline from datasets import load_dataset import gradio as gr import torch from diffusers import DiffusionPipeline pipe_ar = pipeline('text-generation', framework='pt', model='akhooli/ap2023', tokenizer='akhooli/ap2023') pipe_en = pipeline("text-generation", model="ismaelfaro/gpt2-poems.en") pipe_image = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5") pipe_translator = pipeline("translation", model="Helsinki-NLP/opus-mt-ar-en") # Initialize text-to-speech models for Arabic and English # Arabic: text-to-speech synthesiser_arabic = pipeline("text-to-speech", model="facebook/mms-tts-ara") # English: text-to-speech synthesiser_english = pipeline("text-to-speech", model="microsoft/speecht5_tts") embeddings_dataset_english = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation") speaker_embedding_english = torch.tensor(embeddings_dataset_english[7306]["xvector"]).unsqueeze(0) # Generate poem based on language and convert it to audio and image def generate_poem(selected_language, text): if selected_language == "English": poem = generate_poem_english(text) #retrun the generated poem from the generate_poem_english function sampling_rate, audio_data = text_to_speech_english(poem) #return the audio from the text_to_speech_english function image = generate_image_from_poem(poem) #return the image from the generate_image_from_poem function elif selected_language == "Arabic": poem = generate_poem_arabic(text) #retrun the generated poem from the generate_poem_arabic function sampling_rate, audio_data = text_to_speech_arabic(poem) #return the audio from the text_to_speech_arabic function translated_text = translate_arabic_to_english(poem) #return the translated poem from arabic to englsih, using translate_arabic_to_english function image = generate_image_from_poem(translated_text) #return the image from the generate_image_from_poem function return poem, (sampling_rate, audio_data), image # Poem generation for Arabic def generate_poem_arabic(text): generated_text = pipe_ar(text, do_sample=True, max_length=96, top_k=50, top_p=1.0, temperature=1.0, num_return_sequences=1, no_repeat_ngram_size = 3, return_full_text=True)[0]["generated_text"] clean_text = generated_text.replace("-", "") #To get rid of the dashs generated by the model. return clean_text # Poem generation for English def generate_poem_english(text): generated_text = pipe_en(text, do_sample=True, max_length=50)[0]['generated_text'] clean_text = generated_text.replace("-", "") # Remove dashes generated by the model clean_text = clean_text.replace("\\n", " ") # Replace newlines with a space return clean_text # Text-to-speech conversion for Arabic def text_to_speech_arabic(text): speech = synthesiser_arabic(text) audio_data = speech["audio"][0] # Flatten to 1D sampling_rate = speech["sampling_rate"] return (sampling_rate, audio_data) # Text-to-speech conversion for English def text_to_speech_english(text): speech = synthesiser_english(text, forward_params={"speaker_embeddings": speaker_embedding_english}) audio_data = speech["audio"] sampling_rate = speech["sampling_rate"] return (sampling_rate, audio_data) #Image Function def generate_image_from_poem(poem_text): image = pipe_image(poem_text).images[0] return image #Translation Function from Arabic to English def translate_arabic_to_english(text): translated_text = pipe_translator(text)[0]['translation_text'] return translated_text custom_css = """ body { background-color: #f4f4f9; color: #333; } .gradio-container { border-radius: 10px; box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1); background-color: #fff; } label { color: #4A90E2; font-weight: bold; } input[type="text"], textarea { border: 1px solid #4A90E2; } textarea { height: 150px; } button { background-color: #4A90E2; color: #fff; border-radius: 5px; cursor: pointer; } button:hover { background-color: #357ABD; } .dropdown { border: 1px solid #4A90E2; border-radius: 4px; } """ #First parameter is for the dropdown menu, and the second parameter is for the starter of the poem examples = [["Arabic", "الورود تتفتح في فصل الربيع"]] my_model = gr.Interface( fn=generate_poem, #The primary function that will recives the inputs (language and the starter of the poem) inputs=[ gr.Dropdown(["English", "Arabic"], label="Select Language"), #Dropdown menu to select the language, either "English" or "Arabic" for the poem gr.Textbox(label="Enter a sentence")], #Textbox where the user will input a sentence or phrase to generate the poem (starter of the peom) outputs=[ gr.Textbox(label="Generated Poem", lines=10), # Textbox to display the generated poem gr.Audio(label="Generated Audio", type="numpy"), #Audio output for the generated poem gr.Image(label="Generated Image")], #Display an image generated from the starter of the peom examples=examples, #Predefined examples to guide the user how to use the interface css=custom_css #Applying CSS Custeom ) my_model.launch() if __name__ == "__main__": app.run(host="0.0.0.0", port=7860)