Hi everyone,
I’m trying to run Bark locally with the Gradio UI, but I keep running into an issue: the process starts, downloads the models, shows the progress bars, but when it’s supposed to generate the audio, it freezes and Gradio throws errors.

Setup
OS: Windows 11 (build 26100)
Python: 3.10.18 (Anaconda)
GPU: NVIDIA GeForce RTX 4060 Laptop GPU
CUDA: detected correctly (torch.cuda.is_available() == True)
Torch: 2.5.1+cu121
Torchaudio: 2.5.1+cu121
Torchvision: 0.20.1+cu121
Gradio: 3.37.0
Gradio_client: 0.2.10
Websockets: 10.4
Huggingface_hub: 0.35.3
Bark: installed via pip, version not clear (version not reported).

Running on local URL: http://127.0.0.1:7860
IMPORTANT: You are using gradio version 3.37.0, however version 4.44.1 is available, please upgrade.

C:\Users\indi8\anaconda3\envs\bark\lib\site-packages\bark\generation.py:175: FutureWarning: torch.cuda.amp.autocast(args...) is deprecated.
100%|██████████████████████████████████████████| 132/132 [00:03<00:00, 41.10it/s]
100%|██████████████████████████████████████████| 7/7 [00:07<00:00, 1.01s/it]

at this point it freezes and Gradio shows errors

Question

Is this a problem with Gradio / gradio_client version mismatch or with Bark not being fully compatible with Torch 2.5.1?
Could you please share which exact versions are used in the official Bark Hugging Face Space (where everything works), so I can replicate the same environment locally?
Thanks a lot 🙏