--disable_separation

[RobUmf]

def CreateProcessAudio(process_data) -> str:
# Set processing audio to cache file
process_data.process_data_paths.processing_audio_path = os.path.join(
process_data.process_data_paths.cache_folder_path, process_data.basename + ".wav"
)
os_helper.create_folder(process_data.process_data_paths.cache_folder_path)

# --- LOGIC FIX START ---
# Only run the expensive separation if the user actually requested it!
if settings.use_separated_vocal:
    audio_separation_folder_path = separate_vocal_from_audio(
        process_data.process_data_paths.cache_folder_path,
        process_data.process_data_paths.audio_output_file_path,
        settings.use_separated_vocal,
        settings.create_karaoke,
        settings.pytorch_device,
        settings.demucs_model,
        settings.skip_cache_vocal_separation
    )
    process_data.process_data_paths.vocals_audio_file_path = os.path.join(audio_separation_folder_path, "vocals.wav")
    process_data.process_data_paths.instrumental_audio_file_path = os.path.join(audio_separation_folder_path, "no_vocals.wav")
    
    input_path = process_data.process_data_paths.vocals_audio_file_path

else:
    # BYPASS MODE: Use the original file directly
    print(f"{ULTRASINGER_HEAD} {gold_highlighted('Bypass Mode:')} {cyan_highlighted('Skipping Demucs separation.')}")
    
    # We point everything to the original file to prevent crashes
    input_path = process_data.process_data_paths.audio_output_file_path
    process_data.process_data_paths.vocals_audio_file_path = input_path
    process_data.process_data_paths.instrumental_audio_file_path = input_path
# --- LOGIC FIX END ---

# Denoise vocal audio
denoised_output_path = os.path.join(
    process_data.process_data_paths.cache_folder_path, process_data.basename + "_denoised.wav"
)

# We pass the chosen input_path (either separated vocal OR original file)
denoise_vocal_audio(input_path, denoised_output_path, settings.skip_cache_denoise_vocal_audio)

# Convert to mono audio
mono_output_path = os.path.join(
    process_data.process_data_paths.cache_folder_path, process_data.basename + "_mono.wav"
)
convert_audio_to_mono_wav(denoised_output_path, mono_output_path)

# Mute silence sections
mute_output_path = os.path.join(
    process_data.process_data_paths.cache_folder_path, process_data.basename + "_mute.wav"
)
mute_no_singing_parts(mono_output_path, mute_output_path)

# Define the audio file to process
return mute_output_path

[RobUmf]

above around 447 Need to activate a flag?
UltraSinger.py
I'm not going to touch it. Maybe my copy after making a copy of original code.
A go around from separation to creating files for vocal.wav
it is already separated

Currently, the code has a "Fake" flag: --disable_separation. It sets settings.use_separated_vocal = False, but the CreateProcessAudio function ignores this and runs the Demucs downloader anyway.

We don't need to invent a new name; we just need to make the code respect the existing flag.

Here is how to modify UltraSinger.py so that using --disable_separation (or your hard-coded settings) actually skips the download logic entirely.

The Fix: Update CreateProcessAudio
In your UltraSinger.py file, find the CreateProcessAudio function (around line 359) and replace the entire function with this corrected version.

This version adds a simple if check. If use_separated_vocal is False, it executes your "Surgical Bypass" automatically.

[RobUmf]

closer adding
"cache_override_path=", # <--- Add this line within long = [
and around 708
elif opt in ("--cache_override_path"): # <--- ADD THIS BLOCK
settings.cache_override_path = arg

[rakuri255]

@RobUmf just make an PR and lets discuss there

[RobUmf]

Where is there?
I.A fast is being stupid keeps trying to get me to make another fork

[rakuri255]

Yes, you'll need to fork it first. Just commit your changes in an branch there and open a PR. Please use separate branches for different features and try to keep the PRs small and feature wise. Also, please avoid dumping large amounts of AI slop.