import gradio as gr
from PIL import Image, ImageDraw, ImageFont
import scipy.io.wavfile as wavfile
from transformers import pipeline

narrator = pipeline("text-to-speech", model="kakao-enterprise/vits-ljs")
object_detector = pipeline("object-detection", model="facebook/detr-resnet-50")

def generate_audio(text):
    # Generate the narrated text
    narrated_text = narrator(text)
    # Save the audio to a WAV file
    wavfile.write("output.wav", rate=narrated_text["sampling_rate"], data=narrated_text["audio"][0])
    return "output.wav"

def read_objects(detection_objects):
    object_counts = {}
    for detection in detection_objects:
        label = detection['label']
        object_counts[label] = object_counts.get(label, 0) + 1
    response = "This picture contains"
    labels = list(object_counts.keys())
    for i, label in enumerate(labels):
        response += f" {object_counts[label]} {label}" + ("s" if object_counts[label] > 1 else "")
        if i < len(labels) - 2:
            response += ","
        elif i == len(labels) - 2:
            response += " and"
    response += "."
    return response

def draw_bounding_boxes(image, detections, font_path=None, font_size=20):
    draw_image = image.copy()
    draw = ImageDraw.Draw(draw_image)
    if font_path:
        font = ImageFont.truetype(font_path, font_size)
    else:
        font = ImageFont.load_default()
    for detection in detections:
        box = detection['box']
        xmin, ymin, xmax, ymax = box['xmin'], box['ymin'], box['xmax'], box['ymax']
        draw.rectangle([(xmin, ymin), (xmax, ymax)], outline="red", width=3)
        label = detection['label']
        score = detection['score']
        text = f"{label} {score:.2f}"
        if font_path:
            text_size = draw.textbbox((xmin, ymin), text, font=font)
        else:
            text_size = draw.textbbox((xmin, ymin), text)
        draw.rectangle([(text_size[0], text_size[1]), (text_size[2], text_size[3])], fill="red")
        draw.text((xmin, ymin), text, fill="white", font=font)
    return draw_image

def detect_object(image):
    raw_image = image
    output = object_detector(raw_image)
    processed_image = draw_bounding_boxes(raw_image, output)
    natural_text = read_objects(output)
    processed_audio = generate_audio(natural_text)
    return processed_image, processed_audio

examples = [
    ["dogs.jpg"]
]

demo = gr.Interface(
    fn=detect_object,
    inputs=[gr.Image(label="Select Image", type="pil")],
    outputs=[
        gr.Image(label="Processed Image", type="pil"),
        gr.Audio(label="Generated Audio")
    ],
    title="Audio Described Object Detector",
    description="This application highlights objects in the provided image and generates an audio description.",
    examples=examples
)

demo.launch()