I am working on a Python script to remove duplicate MP3 files from a main folder and its subfolders. The challenge is that the file names differ, but the audio content is the same. Additionally, I need to ensure that if any of the duplicate files have hotcues set in Rekordbox (indicated by color attributes in the Rekordbox XML file), those files are preserved.

Current Approach:
Compute a hash for the audio content of each MP3 file using pydub.
Parse the Rekordbox XML file to check if a track has hotcues.
Compare hashes to identify duplicates.
Remove duplicates, ensuring that files with hotcues are kept.
Problem:
The script sometimes fails to correctly identify and preserve files with hotcues, leading to unintended deletions.

Requirements:
Correctly identify and remove duplicate MP3 files.
Preserve any file with hotcues as indicated by color attributes in <POSITION_MARK> elements in the Rekordbox XML.

Example of <POSITION_MARK> with hotcue:
xml

<POSITION_MARK Name="" Type="0" Start="0.236" Num="1" Red="69" Green="172" Blue="219"/>

Current Script:
Here is the latest version of the script I am using:

python

import os
import hashlib
from pydub import AudioSegment
import xml.etree.ElementTree as ET

rekordbox_xml_path = '/path/to/data.xml'

def get_audio_hash(file_path):
    audio = AudioSegment.from_file(file_path)
    audio_bytes = audio.raw_data
    return hashlib.md5(audio_bytes).hexdigest()

def has_hotcues(track):
    for mark in track.findall('.//POSITION_MARK'):
        if 'Red' in mark.attrib or 'Green' in mark.attrib or 'Blue' in mark.attrib:
            return True
    return False

def get_track_by_file_path(xml_root, file_path):
    for track in xml_root.findall('.//TRACK'):
        location = track.get('Location')
        if location and os.path.basename(location) == os.path.basename(file_path):
            return track
    return None

def find_and_remove_duplicates(main_folder, xml_path):
    seen_hashes = {}
    duplicates_to_remove = []

    tree = ET.parse(xml_path)
    xml_root = tree.getroot()

    for root, _, files in os.walk(main_folder):
        for file in files:
            if file.lower().endswith(".mp3"):
                file_path = os.path.join(root, file)
                audio_hash = get_audio_hash(file_path)

                if audio_hash in seen_hashes:
                    existing_file_path = seen_hashes[audio_hash]
                    track_existing = get_track_by_file_path(xml_root, existing_file_path)
                    track_new = get_track_by_file_path(xml_root, file_path)

                    if track_existing is not None and has_hotcues(track_existing):
                        print(f"Hotcue found, keeping: {existing_file_path} (removing {file_path})")
                        duplicates_to_remove.append(file_path)
                    elif track_new is not None and has_hotcues(track_new):
                        print(f"Hotcue found, keeping: {file_path} (removing {existing_file_path})")
                        duplicates_to_remove.append(existing_file_path)
                        seen_hashes[audio_hash] = file_path
                    else:
                        print(f"Duplicate found: {file_path} (duplicate of {existing_file_path}), removing...")
                        duplicates_to_remove.append(file_path)
                else:
                    seen_hashes[audio_hash] = file_path

    for duplicate in duplicates_to_remove:
        if os.path.exists(duplicate):
            os.remove(duplicate)
            print(f"Removed: {duplicate}")

main_folder = '/path/to/main_folder'
find_and_remove_duplicates(main_folder, rekordbox_xml_path)

Question:
How can I modify the script to reliably detect and preserve MP3 files with hotcues while removing other duplicates?