import gradio as gr
import librosa
import torch
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor, MBartForConditionalGeneration, MBart50Tokenizer

# Load pre-trained models
model = Wav2Vec2ForCTC.from_pretrained("boumehdi/wav2vec2-large-xlsr-moroccan-darija")
processor = Wav2Vec2Processor.from_pretrained("boumehdi/wav2vec2-large-xlsr-moroccan-darija")

translation_model = MBartForConditionalGeneration.from_pretrained("facebook/mbart-large-50-many-to-many-mmt")
translation_tokenizer = MBart50Tokenizer.from_pretrained("facebook/mbart-large-50-many-to-many-mmt", src_lang="ar_AR")

def transcribe_audio(audio):
    audio_array, sr = librosa.load(audio, sr=16000)
    input_values = processor(audio_array, return_tensors="pt", padding=True).input_values
    logits = model(input_values).logits
    tokens = torch.argmax(logits, axis=-1)
    transcription = processor.decode(tokens[0])
    translation = translate_text(transcription)
    return transcription, translation

def translate_text(text):
    inputs = translation_tokenizer(text, return_tensors="pt")
    translated_tokens = translation_model.generate(**inputs, forced_bos_token_id=translation_tokenizer.lang_code_to_id["en_XX"])
    translated_text = translation_tokenizer.batch_decode(translated_tokens, skip_special_tokens=True)[0]
    return translated_text

with gr.Blocks() as demo:
    gr.Markdown("# Speech-to-Text and Translation")
    audio_input = gr.Audio(type="filepath")
    submit_button = gr.Button("Submit")
    transcription_output = gr.Textbox(label="Transcription")
    translation_output = gr.Textbox(label="Translation")

    submit_button.click(transcribe_audio, inputs=[audio_input], outputs=[transcription_output, translation_output])

demo.launch()