Re-pitch current UltraStar file

[rootvasilij]

I don't understand how to do it.
Do I need to create input and output folders in the root folder first?
What do I do then?
Kind regards

[rootvasilij]

Have following issue:

(.venv) C:\projects\UltraSinger-main\src>py UltraSinger.py -i "input/ultrastar.txt"

[UltraSinger] *****************************
[UltraSinger] UltraSinger Version: 0.0.12-dev4
[UltraSinger] *****************************
[UltraSinger] Checking GPU support for tensorflow and pytorch.
[UltraSinger] tensorflow - using cuda gpu.
[UltraSinger] pytorch - using cuda gpu.
[UltraSinger] Parse ultrastar txt -> input/ultrastar.txt
Traceback (most recent call last):
File "C:\projects\UltraSinger-main\src\UltraSinger.py", line 693, in
main(sys.argv[1:])
File "C:\projects\UltraSinger-main\src\UltraSinger.py", line 573, in main
run()
File "C:\projects\UltraSinger-main\src\UltraSinger.py", line 133, in run
process_data = InitProcessData()
File "C:\projects\UltraSinger-main\src\UltraSinger.py", line 322, in InitProcessData
) = parse_ultrastar_txt(settings.input_file_path, settings.output_folder_path)
File "C:\projects\UltraSinger-main\src\modules\Ultrastar\ultrastar_parser.py", line 91, in parse_ultrastar_txt
ultrastar_class = parse(input_file_path)
File "C:\projects\UltraSinger-main\src\modules\Ultrastar\ultrastar_parser.py", line 26, in parse
with open(input_file, "r", encoding=FILE_ENCODING) as file:
FileNotFoundError: [Errno 2] No such file or directory: 'input/ultrastar.txt'

[agwosdz]

Make sure the file exists. What this function does is take an existing song's text file and re-pitches it. Let's say you have a folder in c:\Random Song, and the txt file of the song is c:\Random Song\thissong.txt, your command would be py Ultrasinger.py -i "c:\Random Song\thissong.txt"; the software would then take all the needed information from that file, and correct the pitch. This function is only used if you need to correct an existing song. If you copy the folder in to the folder that contains the Ultrasinger.py file, the command would be py ultrasinger.py -i ".\foldername\textfilename.txt"

Hope this makes sense.

[rootvasilij]

[agwosdz]

Do you have CUDA Toolkit 11.8 installed?

Install that and see if it changes anything. Make sure to select custom installation and do not install driver and Geforce Experience.

https://developer.nvidia.com/cuda-11-8-0-download-archive?target_os=Windows&target_arch=x86_64&target_version=11&target_type=exe_network

[rootvasilij]

It's running now, i had a free space issue on my hard disc.

Now I have a language issue, i guess.
English songs seem to work.

Now I have a german song and the fault is as follows:
(.venv) C:\UltraSinger2\UltraSinger-main\src>py ultrasinger.py -i "C:\UltraSinger2\xxx\test.txt"

[UltraSinger] *****************************
[UltraSinger] UltraSinger Version: 0.0.11
[UltraSinger] *****************************
[UltraSinger] Checking GPU support for tensorflow and pytorch.
[UltraSinger] tensorflow - using cuda gpu.
[UltraSinger] pytorch - using cuda gpu.
[UltraSinger] re-pitch mode
[UltraSinger] Parse ultrastar txt -> C:\UltraSinger2\xxx\test.txt
Traceback (most recent call last):
File "C:\UltraSinger2\UltraSinger-main\src\ultrasinger.py", line 991, in
main(sys.argv[1:])
File "C:\UltraSinger2\UltraSinger-main\src\ultrasinger.py", line 883, in main
run()
File "C:\UltraSinger2\UltraSinger-main\src\ultrasinger.py", line 303, in run
) = parse_ultrastar_txt()
File "C:\UltraSinger2\UltraSinger-main\src\ultrasinger.py", line 772, in parse_ultrastar_txt
ultrastar_class = ultrastar_parser.parse_ultrastar_txt(
File "C:\UltraSinger2\UltraSinger-main\src\modules\Ultrastar\ultrastar_parser.py", line 15, in parse_ultrastar_txt
txt = file.readlines()
File "C:\Python310\lib\codecs.py", line 322, in decode
(result, consumed) = self._buffer_decode(data, self.errors, final)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xdf in position 2199: invalid continuation byte

[agwosdz]

Now that you have everything installed properly, I would do a new install of UltraSinger. Just "git clone https://github.com/rakuri255/UltraSinger.git" in a separate folder and run the "windows_cuda_gpu.bat" in the install folder to make sure the virtual environment is all set and all dependencies are installed properly. From the error above, it looks like your venv might have gotten messed up a bit.

[rootvasilij]

Now it seems to run, problem was the .txt file was ANSI and I changed it to UTF-8.

[rakuri255]

@rootvasilij Oh. Can you open an issue for that? UltraSinger should handle ANSI somehow

[rootvasilij]

New I have another issue:

[UltraSinger] Creating repitched ultrastar txt -> C:\test\test.txt_repitch.txt
Traceback (most recent call last):
File "C:\UltraSinger3\src\ultrasinger.py", line 693, in
main(sys.argv[1:])
File "C:\UltraSinger3\src\ultrasinger.py", line 573, in main
run()
File "C:\UltraSinger3\src\ultrasinger.py", line 179, in run
accurate_score, simple_score, ultrastar_file_output = CreateUltraStarTxt(process_data)
File "C:\UltraSinger3\src\ultrasinger.py", line 401, in CreateUltraStarTxt
ultrastar_file_output = create_ultrastar_txt_from_midi_segments(
File "C:\UltraSinger3\src\modules\Ultrastar\coverter\ultrastar_txt_converter.py", line 47, in create_ultrastar_txt_from_midi_segments
create_repitched_txt_from_ultrastar_data(
File "C:\UltraSinger3\src\modules\Ultrastar\ultrastar_writer.py", line 197, in create_repitched_txt_from_ultrastar_data
parts[6] = str(note_numbers[i])
IndexError: list index out of range

[rakuri255]

@rootvasilij You found an bug :). Can you open an issue?

[rootvasilij]

I tried it with a german song via youtube:
(.venv) C:\UltraSinger3\src>py ultrasinger.py -i https://www.youtube.com/watch?v=TSkVVVBS9k8

This is what I got at the end:
[UltraSinger] Loading whisper with model large-v2 and cuda as worker
Requested float16 compute type, but the target device or backend do not support efficient float16 computation.
[UltraSinger] Your GPU does not support efficient float16 computation; run UltraSinger with '--whisper_compute_type int8'
Traceback (most recent call last):
File "C:\UltraSinger3\src\ultrasinger.py", line 693, in
main(sys.argv[1:])
File "C:\UltraSinger3\src\ultrasinger.py", line 573, in main
run()
File "C:\UltraSinger3\src\ultrasinger.py", line 147, in run
TranscribeAudio(process_data)
File "C:\UltraSinger3\src\ultrasinger.py", line 353, in TranscribeAudio
transcription_result = transcribe_audio(process_data.process_data_paths.cache_folder_path,
File "C:\UltraSinger3\src\ultrasinger.py", line 474, in transcribe_audio
transcription_result = transcribe_with_whisper(
File "C:\UltraSinger3\src\modules\Speech_Recognition\Whisper.py", line 104, in transcribe_with_whisper
raise value_error
File "C:\UltraSinger3\src\modules\Speech_Recognition\Whisper.py", line 51, in transcribe_with_whisper
loaded_whisper_model = whisperx.load_model(
File "C:\UltraSinger3.venv\lib\site-packages\whisperx\asr.py", line 288, in load_model
model = model or WhisperModel(whisper_arch,
File "C:\UltraSinger3.venv\lib\site-packages\faster_whisper\transcribe.py", line 133, in init
self.model = ctranslate2.models.Whisper(
ValueError: Requested float16 compute type, but the target device or backend do not support efficient float16 computation.

It seems that I need to add german language first.
I don't get how to do it.

[rakuri255]

its because of your GPU. It cant handle float16. Add this to your call --whisper_compute_type int8

[rootvasilij]

I guess my hardware is too old. I will try it on another pc.

[agwosdz]

Did py ultrasinger.py -i https://www.youtube.com/watch?v=TSkVVVBS9k8 --whisper_compute_type int8 not work for you?

[agwosdz]

Hi, language is not the issue here. As the script tells you, your GPU is not strong enough to do full computation, and you need to add --whisper_compute_type int8 to your command.

py ultrasinger.py -i https://www.youtube.com/watch?v=TSkVVVBS9k8 --whisper_compute_type int8

This should work for your Lied :),

Alternatively, you can start with a fresh install of Ultrasinger and use the windows_CPU.bat from the install/CPU folder to run UltraSinger on your processor and not graphics card. It will take longer and might not be as accurate as on the GPU, but better than nothing (if all fails).

Hope this helps