app.py

import gradio as gr
import whisper
import torch
import os
from pydub import AudioSegment
from transformers import pipeline
from faster_whisper import WhisperModel  # Import faster-whisper

# Mapping of model names to Whisper model sizes
MODELS = {
    "Tiny (Fastest)": "tiny",
    "Base (Faster)": "base",
    "Small (Balanced)": "small",
    "Medium (Accurate)": "medium",
    "Large (Most Accurate)": "large",
    "Systran Faster Whisper Large v3": "Systran/faster-whisper-large-v3"  # Add the new model
}

# Fine-tuned models for specific languages
FINE_TUNED_MODELS = {
    "Tamil": {
        "model": "vasista22/whisper-tamil-medium",
        "language": "ta"
    },
    # Add more fine-tuned models for other languages here
}

# Mapping of full language names to language codes
LANGUAGE_NAME_TO_CODE = {
    "Auto Detect": "Auto Detect",
    "English": "en",
    "Chinese": "zh",
    "German": "de",
    "Spanish": "es",
    "Russian": "ru",
    "Korean": "ko",
    "French": "fr",
    "Japanese": "ja",
    "Portuguese": "pt",
    "Turkish": "tr",
    "Polish": "pl",
    "Catalan": "ca",
    "Dutch": "nl",
    "Arabic": "ar",
    "Swedish": "sv",
    "Italian": "it",
    "Indonesian": "id",
    "Hindi": "hi",
    "Finnish": "fi",
    "Vietnamese": "vi",
    "Hebrew": "he",
    "Ukrainian": "uk",
    "Greek": "el",
    "Malay": "ms",
    "Czech": "cs",
    "Romanian": "ro",
    "Danish": "da",
    "Hungarian": "hu",
    "Tamil": "ta",
    "Norwegian": "no",
    "Thai": "th",
    "Urdu": "ur",
    "Croatian": "hr",
    "Bulgarian": "bg",
    "Lithuanian": "lt",
    "Latin": "la",
    "Maori": "mi",
    "Malayalam": "ml",
    "Welsh": "cy",
    "Slovak": "sk",
    "Telugu": "te",
    "Persian": "fa",
    "Latvian": "lv",
    "Bengali": "bn",
    "Serbian": "sr",
    "Azerbaijani": "az",
    "Slovenian": "sl",
    "Kannada": "kn",
    "Estonian": "et",
    "Macedonian": "mk",
    "Breton": "br",
    "Basque": "eu",
    "Icelandic": "is",
    "Armenian": "hy",
    "Nepali": "ne",
    "Mongolian": "mn",
    "Bosnian": "bs",
    "Kazakh": "kk",
    "Albanian": "sq",
    "Swahili": "sw",
    "Galician": "gl",
    "Marathi": "mr",
    "Punjabi": "pa",
    "Sinhala": "si",  # Sinhala support
    "Khmer": "km",
    "Shona": "sn",
    "Yoruba": "yo",
    "Somali": "so",
    "Afrikaans": "af",
    "Occitan": "oc",
    "Georgian": "ka",
    "Belarusian": "be",
    "Tajik": "tg",
    "Sindhi": "sd",
    "Gujarati": "gu",
    "Amharic": "am",
    "Yiddish": "yi",
    "Lao": "lo",
    "Uzbek": "uz",
    "Faroese": "fo",
    "Haitian Creole": "ht",
    "Pashto": "ps",
    "Turkmen": "tk",
    "Nynorsk": "nn",
    "Maltese": "mt",
    "Sanskrit": "sa",
    "Luxembourgish": "lb",
    "Burmese": "my",
    "Tibetan": "bo",
    "Tagalog": "tl",
    "Malagasy": "mg",
    "Assamese": "as",
    "Tatar": "tt",
    "Hawaiian": "haw",
    "Lingala": "ln",
    "Hausa": "ha",
    "Bashkir": "ba",
    "Javanese": "jw",
    "Sundanese": "su",
}

# Reverse mapping of language codes to full language names
CODE_TO_LANGUAGE_NAME = {v: k for k, v in LANGUAGE_NAME_TO_CODE.items()}

# Device and compute type for faster-whisper
device, torch_dtype = ("cuda", "float32") if torch.cuda.is_available() else ("cpu", "int8")

def detect_language(audio_file):
    """Detect the language of the audio file."""
    # Load the Whisper model (use "base" for faster detection)
    model = whisper.load_model("base")
    
    # Convert audio to 16kHz mono for better compatibility with Whisper
    audio = AudioSegment.from_file(audio_file)
    audio = audio.set_frame_rate(16000).set_channels(1)
    processed_audio_path = "processed_audio.wav"
    audio.export(processed_audio_path, format="wav")
    
    # Detect the language
    result = model.transcribe(processed_audio_path, task="detect_language", fp16=False)
    detected_language_code = result.get("language", "unknown")
    
    # Get the full language name from the code
    detected_language = CODE_TO_LANGUAGE_NAME.get(detected_language_code, "Unknown Language")
    
    # Clean up processed audio file
    os.remove(processed_audio_path)
    
    return f"Detected Language: {detected_language}"

def transcribe_audio(audio_file, language="Auto Detect", model_size="Base (Faster)"):
    """Transcribe the audio file."""
    # Convert audio to 16kHz mono for better compatibility
    audio = AudioSegment.from_file(audio_file)
    audio = audio.set_frame_rate(16000).set_channels(1)
    processed_audio_path = "processed_audio.wav"
    audio.export(processed_audio_path, format="wav")
    
    # Load the appropriate model
    if language in FINE_TUNED_MODELS:
        # Use the fine-tuned Whisper model for the selected language
        device = "cuda:0" if torch.cuda.is_available() else "cpu"
        transcribe = pipeline(
            task="automatic-speech-recognition",
            model=FINE_TUNED_MODELS[language]["model"],
            chunk_length_s=30,
            device=device
        )
        transcribe.model.config.forced_decoder_ids = transcribe.tokenizer.get_decoder_prompt_ids(
            language=FINE_TUNED_MODELS[language]["language"],
            task="transcribe"
        )
        result = transcribe(processed_audio_path)
        transcription = result["text"]
        detected_language = language
    else:
        # Use the selected Whisper model
        if model_size == "Systran Faster Whisper Large v3":
            # Use faster-whisper for the Systran model
            model = WhisperModel(MODELS[model_size], device=device, compute_type=torch_dtype)
            segments, info = model.transcribe(
                processed_audio_path,
                task="transcribe",
                word_timestamps=True,
                repetition_penalty=1.1,
                temperature=[0.0, 0.1, 0.2, 0.3, 0.4, 0.6, 0.8, 1.0],
            )
            transcription = " ".join([segment.text for segment in segments])
            detected_language_code = info.language
            detected_language = CODE_TO_LANGUAGE_NAME.get(detected_language_code, "Unknown Language")
        else:
            # Use the standard Whisper model
            model = whisper.load_model(MODELS[model_size])
            
            # Transcribe the audio
            if language == "Auto Detect":
                result = model.transcribe(processed_audio_path, fp16=False)  # Auto-detect language
                detected_language_code = result.get("language", "unknown")
                detected_language = CODE_TO_LANGUAGE_NAME.get(detected_language_code, "Unknown Language")
            else:
                language_code = LANGUAGE_NAME_TO_CODE.get(language, "en")  # Default to English if not found
                result = model.transcribe(processed_audio_path, language=language_code, fp16=False)
                detected_language = language
            
            transcription = result["text"]
    
    # Clean up processed audio file
    os.remove(processed_audio_path)
    
    # Return transcription and detected language
    return f"Detected Language: {detected_language}\n\nTranscription:\n{transcription}"

# Define the Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("# Audio Transcription and Language Detection")
    
    with gr.Tab("Detect Language"):
        gr.Markdown("Upload an audio file to detect its language.")
        detect_audio_input = gr.Audio(type="filepath", label="Upload Audio File")
        detect_language_output = gr.Textbox(label="Detected Language")
        detect_button = gr.Button("Detect Language")
    
    with gr.Tab("Transcribe Audio"):
        gr.Markdown("Upload an audio file, select a language (or choose 'Auto Detect'), and choose a model for transcription.")
        transcribe_audio_input = gr.Audio(type="filepath", label="Upload Audio File")
        language_dropdown = gr.Dropdown(
            choices=list(LANGUAGE_NAME_TO_CODE.keys()),  # Full language names
            label="Select Language",
            value="Auto Detect"
        )
        model_dropdown = gr.Dropdown(
            choices=list(MODELS.keys()),  # Model options
            label="Select Model",
            value="Base (Faster)",  # Default to "Base" model
            interactive=True  # Allow model selection by default
        )
        transcribe_output = gr.Textbox(label="Transcription and Detected Language")
        transcribe_button = gr.Button("Transcribe Audio")
    
    # Link buttons to functions
    detect_button.click(detect_language, inputs=detect_audio_input, outputs=detect_language_output)
    transcribe_button.click(transcribe_audio, inputs=[transcribe_audio_input, language_dropdown, model_dropdown], outputs=transcribe_output)

# Launch the Gradio interface
demo.launch()