app.py

import subprocess
import sys

# Force upgrade gradio
subprocess.check_call([sys.executable, "-m", "pip", "install", "--upgrade", "gradio>=4.44.0"])

from transformers import (
    pipeline,
    WhisperForConditionalGeneration,
    AutoTokenizer,
    WhisperFeatureExtractor,
    GenerationConfig
)
import gradio as gr
import traceback

print("🔄 Starting ASR application...")

# Global ASR pipeline
asr = None

def load_asr_pipeline():
    global asr
    try:
        print("🔧 Loading ASR model and configuration...")

        model_id = "amedcj/whisper-kurmanji"

        # Load generation config and remove forced_decoder_ids
        gen_config = GenerationConfig.from_pretrained(model_id)
        gen_config.forced_decoder_ids = None

        # Load model and attach config
        model = WhisperForConditionalGeneration.from_pretrained(model_id)
        model.generation_config = gen_config

        # Load tokenizer and feature extractor
        tokenizer = AutoTokenizer.from_pretrained(model_id)
        feature_extractor = WhisperFeatureExtractor.from_pretrained(model_id)

        # Assemble pipeline
        asr = pipeline(
            "automatic-speech-recognition",
            model=model,
            tokenizer=tokenizer,
            feature_extractor=feature_extractor,
            device=-1  # Use CPU
        )
        print("✅ ASR pipeline loaded successfully.")
    
    except Exception as e:
        print(f"❌ Failed to load ASR model: {e}")
        traceback.print_exc()
        asr = None

# Load on startup
load_asr_pipeline()

def transcribe(audio_file):
    print("🎙️ Transcribe function triggered.")
    
    if not audio_file:
        msg = "⚠️ Please upload an audio file."
        print(msg)
        return msg
    
    if asr is None:
        msg = "❌ ASR model not loaded properly."
        print(msg)
        return msg

    try:
        print(f"🔍 Transcribing: {audio_file}")
        result = asr(audio_file)
        print("✅ Transcription complete.")
        return result["text"]
    except Exception as e:
        error_msg = f"❌ Error during transcription: {str(e)}"
        print(error_msg)
        traceback.print_exc()
        return error_msg

print("🚧 Building Gradio interface...")

interface = gr.Interface(
    fn=transcribe,
    inputs=gr.Audio(type="filepath", label="🎤 Dengê Kurmancî barkirin / Upload Kurdish Audio"),
    outputs=gr.Textbox(label="📝 Nivîsandin / Transcription"),
    title="Dengê Kurmancî bo Nivîsandin – Kurdish ASR",
    description="Dengê Kurmancî barkirin û nivîsa wê bibînin. / Upload Kurdish audio and get the transcription.",
    submit_btn="Bişîne",
    clear_btn="Paqij bike",
    examples=[
        ["samples/kurmanji_sample.wav"]
    ]
)

print("🚀 Launching ASR interface...")

if __name__ == "__main__":
    interface.launch()