Consistently getting CUFFT_INTERNAL_ERROR when using any voice other than "random"

[rancortooth]

Hopefully I'm just doing something stupid and that's why I'm seeing this error. Here's my setup:

• Windows 11, running in WSL
• NVIDIA GeForce RTX 4070, 4GB VRAM
• NVIDIA-SMI 535.103, Driver Version: 537.13, CUDA Version: 12.2

Everything installed just fine using conda. If I run this command, it works fantastic:

python tortoise/do_tts.py --text "I'm going to speak this" --voice random --preset fast

If I use anything other than "random" for the voice (e.g. "train_mouse") , I get CUFFT_INTERNAL_ERROR very consistently. Any help is very much appreciated! Thank you!

Traceback (most recent call last):
  File "/home/rancortooth/ai/tortoise-tts/tortoise/do_tts.py", line 41, in <module>
    gen, dbg_state = tts.tts_with_preset(args.text, k=args.candidates, voice_samples=voice_samples, conditioning_latents=conditioning_latents,
  File "/home/rancortooth/ai/tortoise-tts/tortoise/api.py", line 346, in tts_with_preset
    return self.tts(text, **settings)
  File "/home/rancortooth/ai/tortoise-tts/tortoise/api.py", line 409, in tts
    auto_conditioning, diffusion_conditioning, auto_conds, _ = self.get_conditioning_latents(voice_samples, return_mels=True)
  File "/home/rancortooth/ai/tortoise-tts/tortoise/api.py", line 291, in get_conditioning_latents
    auto_conds.append(format_conditioning(vs, device=self.device))
  File "/home/rancortooth/ai/tortoise-tts/tortoise/api.py", line 113, in format_conditioning
    mel_clip = TorchMelSpectrogram()(clip.unsqueeze(0)).squeeze(0)
  File "/home/rancortooth/miniconda3/envs/tortoise/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/rancortooth/miniconda3/envs/tortoise/lib/python3.9/site-packages/TorToiSe-2.7.0-py3.9.egg/tortoise/models/arch_util.py", line 325, in forward
    mel = self.mel_stft(inp)
  File "/home/rancortooth/miniconda3/envs/tortoise/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/rancortooth/miniconda3/envs/tortoise/lib/python3.9/site-packages/torchaudio/transforms/_transforms.py", line 650, in forward
    specgram = self.spectrogram(waveform)
  File "/home/rancortooth/miniconda3/envs/tortoise/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/rancortooth/miniconda3/envs/tortoise/lib/python3.9/site-packages/torchaudio/transforms/_transforms.py", line 110, in forward
    return F.spectrogram(
  File "/home/rancortooth/miniconda3/envs/tortoise/lib/python3.9/site-packages/torchaudio/functional/functional.py", line 126, in spectrogram
    spec_f = torch.stft(
  File "/home/rancortooth/miniconda3/envs/tortoise/lib/python3.9/site-packages/torch/functional.py", line 641, in stft
    return _VF.stft(input, n_fft, hop_length, win_length, window,  # type: ignore[attr-defined]
RuntimeError: cuFFT error: CUFFT_INTERNAL_ERROR

[DanHues]

I am having this same exact issue, did you ever find a solution?

Specs

• OS: Ubuntu 22.04.3 LTS
• Python Version: 3.11.5
• Conda Environment: Yes
• CUDA Version 12.2
• Hardware: 4060 8gb VRAM Laptop

[StoryHack]

I had the same issue. Eventually, I changed how I was installing tortoise. rather than using the command:

conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia

from the installation instructions, I used:

conda install pytorch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 pytorch-cuda=11.8 -c pytorch -c nvidia

and was finally able to use the built in voices.