Tortoise and RVC don't work.

[edbartz]

I have tried installing this 3 times. Following each step in the instructions. Then attempting the test you show. It never works. One thing I notice, is it says RVC is not installed. I downloaded the whl fresh for each install, and followed those steps. When I test things, the windows tts always works. When I test tortoise, with use s2s unchecked, I get the following:
C:\Users\xxx\Desktop\TTS\audiobook_maker>call venv\Scripts\activate
[2024-10-22 10:38:19,263] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)
test.c
LINK : fatal error LNK1181: cannot open input file 'aio.lib'
RVC not installed, received error: CUDA_HOME does not exist, unable to compile CUDA op(s)
Loading TorToiSe... (AR: None, diffusion: , vocoder: )
Hardware acceleration found: cuda
use_deepspeed api_debug True
C:\Users\xxx\Desktop\TTS\audiobook_maker\venv\Lib\site-packages\huggingface_hub\file_download.py:797: FutureWarning: resume_download is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use force_download=True.
warnings.warn(
C:\Users\xxx\Desktop\TTS\audiobook_maker\venv\Lib\site-packages\torch\nn\utils\weight_norm.py:28: UserWarning: torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.
warnings.warn("torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.")
Loading tokenizer JSON: C:\Users\xxx\Desktop\TTS\audiobook_maker\venv\Lib\site-packages\tortoise../tortoise/data/tokenizer.json
Loaded tokenizer
Downloading autoregressive.pth from https://huggingface.co/jbetker/tortoise-tts-v2/resolve/main/.models/autoregressive.pth...
100% |########################################################################|
Done.
Loading autoregressive model: C:\Users\xxx\Desktop\TTS\audiobook_maker\models\tortoise\autoregressive.pth
test.c
LINK : fatal error LNK1181: cannot open input file 'aio.lib'
Error calling Python override of QThread::run(): Traceback (most recent call last):
File "C:\Users\xxx\Desktop\TTS\audiobook_maker\src\controller.py", line 37, in run
self.function(self.directory_path, self.is_continue, self.report_progress, self.sentence_generated_callback, self.should_stop)
File "C:\Users\xxx\Desktop\TTS\audiobook_maker\src\model.py", line 260, in generate_audio_for_sentence_threaded
self.load_selected_tts_engine(tts_engine_name, speaker_id, **speaker_settings)
File "C:\Users\xxx\Desktop\TTS\audiobook_maker\src\model.py", line 205, in load_selected_tts_engine
self.tts_engine = tts_engines.load_tts_engine(chosen_tts_engine, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\xxx\Desktop\TTS\audiobook_maker\src\tts_engines.py", line 80, in load_tts_engine
raise e
File "C:\Users\xxx\Desktop\TTS\audiobook_maker\src\tts_engines.py", line 72, in load_tts_engine
return load_with_tortoise(**kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\xxx\Desktop\TTS\audiobook_maker\src\tts_engines.py", line 118, in load_with_tortoise
tts = load_tortoise_engine(
^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\xxx\Desktop\TTS\audiobook_maker\venv\Lib\site-packages\tortoise_tts_api\inference\load.py", line 38, in load_tts
tts = TorToise_TTS(
^^^^^^^^^^^^^
File "C:\Users\xxx\Desktop\TTS\audiobook_maker\venv\Lib\site-packages\tortoise\api.py", line 308, in init
self.load_autoregressive_model(autoregressive_model_path)
File "C:\Users\xxx\Desktop\TTS\audiobook_maker\venv\Lib\site-packages\tortoise\api.py", line 393, in load_autoregressive_model
self.autoregressive.post_init_gpt2_config(use_deepspeed=self.use_deepspeed, kv_cache=self.use_kv_cache)
File "C:\Users\xxx\Desktop\TTS\audiobook_maker\venv\Lib\site-packages\tortoise\models\autoregressive.py", line 387, in post_init_gpt2_config
import deepspeed
File "C:\Users\xxx\Desktop\TTS\audiobook_maker\venv\Lib\site-packages\deepspeed_init_.py", line 25, in
from . import ops
File "C:\Users\xxx\Desktop\TTS\audiobook_maker\venv\Lib\site-packages\deepspeed\ops_init_.py", line 15, in
from ..git_version_info import compatible_ops as compatible_ops
File "C:\Users\xxx\Desktop\TTS\audiobook_maker\venv\Lib\site-packages\deepspeed\git_version_info.py", line 29, in
op_compatible = builder.is_compatible()
^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\xxx\Desktop\TTS\audiobook_maker\venv\Lib\site-packages\deepspeed\ops\op_builder\fp_quantizer.py", line 35, in is_compatible
sys_cuda_major, _ = installed_cuda_version()
^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\xxx\Desktop\TTS\audiobook_maker\venv\Lib\site-packages\deepspeed\ops\op_builder\builder.py", line 51, in installed_cuda_version
raise MissingCUDAException("CUDA_HOME does not exist, unable to compile CUDA op(s)")
deepspeed.ops.op_builder.builder.MissingCUDAException: CUDA_HOME does not exist, unable to compile CUDA op(s)

[deffcolony]

I made a Audiobook-maker launcher which uses miniconda3 to install all requirements

You can take a look here see if it works for you: #64

Just make sure to run it from the audiobook_maker folder that you already cloned

[JarodMica]

Hey @edbartz ,

I believe it comes down to this here: https://developer.nvidia.com/cuda-12-1-0-download-archive

Could you install this and see if it resolves your issue?

I overlooked this stage as I've had it pre installed on my computer for awhile

[edbartz]

Interesting. That fixed it. Which I don't quite understand, as I had cuda_12.5.0_555.85_windows.exe installed. Which "appears" to be a newer version, but thanks. it seems to be working again. One minor question, using the same tortoise and rvc voice as with the old maker, the rvc result is much deeper voiced. No clue why, but interesting.... Thanks again. Keep up the great work!!! I love you stuff. Now to add in the word replacement stuff I added from the 2.0 maker.

[JarodMica]

Interesting. That fixed it. Which I don't quite understand, as I had cuda_12.5.0_555.85_windows.exe installed. Which "appears" to be a newer version, but thanks. it seems to be working again. Thanks again. Keep up the great work!!! I love you stuff. Now to add in the word replacement stuff I added from the 2.0 maker.

I'm not exactly sure on why RVC require cuda home, but deepspeed I could expect it I guess. I think it's because you need cuda toolkit at 12.1, thought, I thought it should work if you had it at 12.5 as well.

One minor question, using the same tortoise and rvc voice as with the old maker, the rvc result is much deeper voiced. No clue why, but interesting....

There's an unfortunate bug that I think the author of the rvc-python library didn't fix, which was fixed in RVC, but when you use sample rate at 48k, it messes up v1 RVC models. So you can bring that down to 0, I'll have to adjust the default values for it in the config

[edbartz]

I will have to see if I can make a different version voice models. Unfortunately, the only way I figured out to make rvc voices is using applio. I would prefer to do it local, but I can't see how to make rvc voices using your programs, so we will see how applio does.

[edbartz]

Jarod,
I found the problem with the deep voices. I looked into how I created the voices, and they were done at 40K, not 48K. When I adjusted the slider accordingly, it worked fine. I will either have to recreate the voices, or mod the code for 40k by default.
Thanks again.
Ed (aka physengineer)

[JarodMica]

The weird thing is that this is was fixed in the main repo of RVC a long time ago, I guess the author I took the rvc API from used older code possibly... But yes! It works just fine with v2/48k models

[edbartz]

I am redoing the voices now, in 48k. I might as well, since I am using a 3090 now, rather than a 3070ti. I can't wait for you do the video about setting up tortoise voices (with training). I tried guessing what I should do with the trained pth file from tortoise, but that failed, so guessing is obviously not my forte, so, I will wait patiently, until you have time.
Ed (aka physengineer)