app.py

import gradio as gr
import whisper
from deep_translator import GoogleTranslator
import nltk
nltk.download('punkt')

def transcribe_audio(audio, model_name):
    model = whisper.load_model(model_name)
    result = model.transcribe(audio)
    return result["text"]

def translate_transcript(transcript_text, target_language, max_chunk_length=5000):
    print("Translating into", target_language)
    translator = GoogleTranslator(source='auto', target=target_language)
    
    # Split content into chunks that attempt to maintain context
    chunks = split_text_into_chunks(transcript_text, max_chunk_length)
    
    translated_chunks = []
    for chunk in chunks:
        # Translate each chunk
        translated_chunks.append(translator.translate(chunk.strip()))
    
    # Join all translated chunks into a single string
    translated_text = ' '.join(translated_chunks)
    
    return translated_text

def split_text_into_chunks(text, max_chunk_length):
    """
    Helper function to split text into chunks that attempt to maintain context.
    """
    # Split text into smaller chunks based on logical points (e.g., pauses, transitions)
    chunks = []
    current_chunk = ""
    words = nltk.word_tokenize(text)
    
    for word in words:
        if len(current_chunk) + len(word) < max_chunk_length:
            current_chunk += word + " "
        else:
            if current_chunk:
                chunks.append(current_chunk.strip())
            current_chunk = word + " "
    
    if current_chunk:
        chunks.append(current_chunk.strip())
    
    return chunks

# Example usage function
def transcribe_and_translate(audio, target_language):
    if not target_language:
        target_language = "English"
    target_language_code = lang_name_to_code[target_language]
    
    # Transcribe audio
    transcript_text = transcribe_audio(audio, model_name="base")
    
    # Translate transcript to the target language
    translated_text = translate_transcript(transcript_text, target_language=target_language_code)
    
    return translated_text

# List of top 10 widely used languages with their codes
top_languages = [
    ("English", "en"),
    ("Chinese", "zh"),
    ("Spanish", "es"),
    ("Hindi", "hi"),
    ("Arabic", "ar"),
    ("Portuguese", "pt"),
    ("Bengali", "bn"),
    ("Russian", "ru"),
    ("Japanese", "ja"),
    ("Punjabi", "pa"),
]
lang_name_to_code = {name: code for name, code in top_languages}

# Gradio interface
demo = gr.Interface(
    fn=transcribe_and_translate,
    inputs=[
        gr.Audio(type="filepath"),
        gr.Dropdown(choices=[lang[0] for lang in top_languages], label="Language")
    ],
    outputs="textbox",
)
demo.launch()