import gradio as gr
from transformers import pipeline
import numpy as np
import os
from huggingface_hub import login
import spaces

# Get token from Space secrets
HF_TOKEN = os.environ.get("HF_TOKEN")
if HF_TOKEN:
    login(token=HF_TOKEN)

# Load model from your private repo
MODEL_ID = "badrex/JASR"  # Change this to match your repo!
transcriber = pipeline("automatic-speech-recognition", model=MODEL_ID)

@spaces.GPU
def transcribe(audio):
    sr, y = audio
    # Convert to mono if stereo
    if y.ndim > 1:
        y = y.mean(axis=1)
    y = y.astype(np.float32)
    y /= np.max(np.abs(y))
    return transcriber({"sampling_rate": sr, "raw": y})["text"]

demo = gr.Interface(
    fn=transcribe,
    inputs=gr.Audio(),
    outputs="text",
    title="JASR 🐐 Dialectal Arabic ASR",
    description="""
        <div class="centered-content">
            <div>
                <p>
                By <a href="https://badrex.github.io/" style="color: #2563eb;">Badr al-Absi</a> with ❤️🤍💚 
                </p>
                <br>
                <p style="font-size: 15px; line-height: 1.8;">
                This is JASR, pronounced Jasir, an automatic speech recognition system optimized for the regional dialects of <i>Jazirat al-Arab</i>, or the Arabian Peninsula. The model is a fine-tune of the speech foundation model <https://huggingface.co/facebook/w2v-bert-2.0" style="color: #FF5349;">w2v-BERT 2.0</a>, a 580M pre-trained speech encoder. 
                <br>                   
                <p style="font-size: 15px; line-height: 1.8;">
                Simply <strong>upload an audio file</strong> 📤 or <strong>record yourself speaking</strong> 🎙️⏺️ to try out the model!
                </p>
            </div>
        </div>
        """,
    examples=examples if examples else None,
    cache_examples=False,  # Disable caching to avoid issues
    flagging_mode=None,
)

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