Merge branch 'main' into CI

# Conflicts:
#	requirements.txt
This commit is contained in:
Vadim committed 2023-06-08 01:43:58 +02:00
commit 2e9d126777
6 files changed
+68 -24

No files matched your search

+37
View File
@@ -44,6 +44,7 @@ This will help me alot to keep this project alive and improve it.
* Type `.venv\Scripts\activate` and press enter.
* You should see now a `(.venv)` in front of your console line.
* Install the requirements with `pip install -r requirements.txt`.
* Install gpu requirements `pip3 install torch==2.0.1+cu117 torchvision==0.15.2+cu117 torchaudio==2.0.2+cu117 --index-url https://download.pytorch.org/whl/cu117`
* Now you can use the UltraSinger source code with `py UltraSinger.py [opt] [mode] [transcription] [pitcher] [extra]`. See [How to use](#how-to-use) for more information.
For more information about Python environments look [here](https://code.visualstudio.com/docs/python/environments#_global-virtual-or-conda-environments).
@@ -54,6 +55,7 @@ As copy:
py -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
pip3 install torch==2.0.1+cu117 torchvision==0.15.2+cu117 torchaudio==2.0.2+cu117 --index-url https://download.pytorch.org/whl/cu117
```
@@ -195,3 +197,38 @@ for low male voices. Accurate is the real tone specified in the txt. I had txt f
singable by humans, but you could still reach the 10k points in the game. The accuracy is important here, because from
this MIDI and sheet are created. And you also want to have accurate files
### Use GPU
With an GPU you can speed up the process and also the quality of the transcription and pitching is better.
You need a cuda device for this to work.
For best use you need to install the cuda driver for your gpu see [here](https://developer.nvidia.com/cuda-downloads).
Install torch with cuda separately in your `venv`. See this [link](https://pytorch.org/get-started/locally/).
Command for `pip`:
```
pip3 install torch~=2.0.1+cu117 torchvision~=0.15.2+cu117 torchaudio~=2.0.2+cu117 --index-url https://download.pytorch.org/whl/cu117
```
When you want to use `conda` instead you need a different installation command. See this [link](https://pytorch.org/get-started/locally/).
#### Info
If somthing crash because of low VRAM than use a smaller model.
Whisper needs more than 8GB VRAM in the `large` model!
But you can force cpu usage with the extra options `--force_whisper_cpu` and `--force_separation_cpu`.
### Use GPU
With an GPU you can speed up the process and also the quality of the transcription and pitching is better.
You need a cuda device for this to work.
If you use MAC than sorry, there is no cuda device for MAC machinges.
#### Info
If somthing crash because of low VRAM than use a smaller model.
Whisper needs more than 8GB VRAM in the `large` model!
But you can force cpu usage with the extra options `--force_whisper_cpu` and `--force_separation_cpu`.
+4 -1
View File
@@ -26,4 +26,7 @@ class Settings:
crepe_step_size = 10
# Device
device = "cpu" # cpu|cuda|gpu
device = 'cpu' # cpu|cuda
force_whisper_cpu = False
force_separation_cpu = False
+11 -4
View File
@@ -437,9 +437,9 @@ def get_unused_song_output_dir(path):
def transcribe_audio(transcribed_data):
"""Docstring"""
if settings.transcriber == "whisper":
device = "cpu" if settings.force_whisper_cpu else settings.device
transcribed_data, language = transcribe_with_whisper(
settings.mono_audio_path, settings.whisper_model, settings.device
)
settings.mono_audio_path, settings.whisper_model, device)
else: # vosk
transcribed_data = transcribe_with_vosk(
settings.mono_audio_path, settings.vosk_model_path
@@ -450,14 +450,15 @@ def transcribe_audio(transcribed_data):
def separate_vocal_from_audio(
basename_without_ext, cache_path, ultrastar_audio_input_path
basename_without_ext, cache_path, ultrastar_audio_input_path
):
"""Docstring"""
audio_separation_path = os.path.join(
cache_path, "separated", "htdemucs", basename_without_ext
)
device = "cpu" if settings.force_separation_cpu else settings.device
if settings.use_separated_vocal or settings.create_karaoke:
separate_audio(ultrastar_audio_input_path, cache_path)
separate_audio(ultrastar_audio_input_path, cache_path, device)
if settings.use_separated_vocal:
vocals_path = os.path.join(audio_separation_path, "vocals.wav")
convert_audio_to_mono_wav(vocals_path, settings.mono_audio_path)
@@ -761,6 +762,8 @@ def init_settings(argv):
"disable_separation=",
"disable_karaoke=",
"create_audio_chunks=",
"force_whisper_cpu=",
"force_separation_cpu="
]
opts, args = getopt.getopt(argv, short, long)
if len(opts) == 0:
@@ -790,6 +793,10 @@ def init_settings(argv):
settings.create_karaoke = not arg
elif opt in ("--create_audio_chunks"):
settings.create_audio_chunks = arg
elif opt in ("--force_whisper_cpu"):
settings.force_whisper_cpu = arg
elif opt in ("--force_separation_cpu"):
settings.force_separation_cpu = arg
if settings.output_file_path == "":
if settings.input_file_path.startswith("https:"):
dirname = os.getcwd()
+3 -3
View File
@@ -19,9 +19,9 @@ def separate_audio(input_file_path: str, output_file: str, device="cpu") -> None
# Model selection?
# -n mdx_q
# -n htdemucs_ft
# todo: -d cpu otherwise it automatically uses gpu
device = "cpu" if device != "cuda" else "cuda"
subprocess.run(
["demucs", "--two-stems=vocals", input_file_path], check=True
["demucs", "-d", device, "--two-stems=vocals", input_file_path]
)
separated_folder = path_join(current_executor_path(), "separated")
move(separated_folder, output_file)
move(separated_folder, output_file)
+8 -13
View File
@@ -1,26 +1,21 @@
"""Docstring"""
import tensorflow as tf
import torch
from modules.console_colors import ULTRASINGER_HEAD, red_highlighted
def get_available_device():
"""Docstring"""
gpus = tf.config.list_physical_devices("GPU")
isCuda = torch.cuda.is_available()
if not gpus:
if not isCuda:
print(
f"{ULTRASINGER_HEAD} There are no GPUs available. Using {red_highlighted('cpu')}."
f"{ULTRASINGER_HEAD} There are no {red_highlighted('cuda')} devices available. -> Using {red_highlighted('cpu')}."
)
return "cpu"
else:
print(f"{ULTRASINGER_HEAD} Using {red_highlighted('cuda')} GPU.")
print(f"{ULTRASINGER_HEAD} Found available GPUs:")
for gpu in gpus:
print(f"Name: {gpu.name}, Type: {gpu.device_type}")
# Todo: Finish this
print(
f"{ULTRASINGER_HEAD} GPU usage are currently in development. Using {red_highlighted('cpu')}"
)
return "cpu"
return "cuda"
+5 -3
View File
@@ -1,11 +1,13 @@
crepe==0.0.13
librosa==0.9.2
pretty_midi==0.2.9
pretty_midi~=0.2.10
pydub==0.25.1
Levenshtein==0.20.9
scipy==1.10.0
Levenshtein~=0.21.0
scipy~=1.10.1
vosk==0.3.44
tensorflow~=2.12.0
#tensorflow-directml-plugin~=0.4.0.dev230202
#cudatoolkit==11.8.0
matplotlib~=3.7.1
yt_dlp~=2023.3.4
numpy~=1.23.5