app.py

import os
import gradio as gr
from faster_whisper import WhisperModel
import google.generativeai as genai
from gtts import gTTS, lang
import tempfile

# Configure Gemini API (replace with your API key or use environment variable)
GEMINI_API_KEY = os.getenv("GEMINI_API_KEY", "YOUR_GEMINI_API_KEY_HERE")
genai.configure(api_key=GEMINI_API_KEY)

# Initialize the faster-whisper model
model_size = "Systran/faster-whisper-large-v3"
whisper_model = WhisperModel(model_size, device="auto", compute_type="float16")

# Function to transcribe audio using faster-whisper
def transcribe_audio(audio_file):
    try:
        segments, info = whisper_model.transcribe(audio_file, beam_size=5)
        transcription = " ".join([segment.text for segment in segments])
        detected_language = info.language
        return transcription, detected_language, None
    except Exception as e:
        return None, None, f"Transcription error: {str(e)}"

# Function to translate text using Gemini API with a magic prompt
def translate_text(text, target_language):
    try:
        model = genai.GenerativeModel("gemini-1.5-flash")
        # Magic prompt to ensure only translated text is returned
        prompt = f"Translate the following text to {target_language} and return only the translated text with no additional explanation or commentary:\n\n{text}"
        response = model.generate_content(prompt)
        translated_text = response.text.strip()
        return translated_text, None
    except Exception as e:
        return None, f"Translation error: {str(e)}"

# Function to convert text to speech using gTTS with full language support
def text_to_speech(text, language):
    try:
        # Get all supported languages from gTTS
        lang_map = lang.tts_langs()
        # Use the language code directly if supported, otherwise default to 'en'
        tts_lang = language.lower() if language.lower() in lang_map else "en"
        tts = gTTS(text=text, lang=tts_lang, slow=False)
        with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as fp:
            tts.save(fp.name)
            return fp.name, None
    except Exception as e:
        return None, f"TTS error: {str(e)}"

# Main function to process audio input and return outputs
def process_audio(audio_file, target_language):
    # Step 1: Transcribe audio
    transcription, detected_language, error = transcribe_audio(audio_file)
    if error:
        return error, None, None, None
    
    # Step 2: Translate transcription
    translated_text, error = translate_text(transcription, target_language)
    if error:
        return error, transcription, None, None
    
    # Step 3: Convert translated text to speech
    # Map target language name to gTTS language code
    lang_map = lang.tts_langs()
    # Convert target_language to lowercase keys as in lang_map
    lang_key = next((k for k, v in lang_map.items() if v.lower() == target_language.lower()), "en")
    audio_output, error = text_to_speech(translated_text, lang_key)
    if error:
        return error, transcription, translated_text, None
    
    return None, transcription, translated_text, audio_output

# Gradio interface
with gr.Blocks(title="AI Audio Translator") as demo:
    gr.Markdown("# AI Audio Translator")
    gr.Markdown("Upload an audio file, select a target language, and get the transcription, translation, and translated audio!")
    
    # Get all supported languages from gTTS
    supported_langs = {v: k for k, v in lang.tts_langs().items()}  # {name: code}
    language_choices = list(supported_langs.keys())  # List of language names
    
    with gr.Row():
        audio_input = gr.Audio(sources=["upload", "microphone"], type="filepath", label="Input Audio")
        target_lang = gr.Dropdown(
            choices=sorted(language_choices),
            value="Spanish",
            label="Target Language"
        )
    
    submit_btn = gr.Button("Translate")
    
    with gr.Row():
        error_output = gr.Textbox(label="Error", visible=True)
        transcription_output = gr.Textbox(label="Transcription")
        translation_output = gr.Textbox(label="Translated Text")
        audio_output = gr.Audio(label="Translated Audio")
    
    submit_btn.click(
        fn=process_audio,
        inputs=[audio_input, target_lang],
        outputs=[error_output, transcription_output, translation_output, audio_output]
    )

# Launch the app
demo.launch()