[BUG] Whishper "transcripton failed" Error

[JoshuaH-072017]

As the title implies, regardless of what option I select or what video I input (whether URL or file), the red error message of "transcription failed" always appears. I've attempted following the manual's instructions regarding this type of error (such as making sure the application's local IP address was inserted in the "docker-compose" file and checking the log files), but no success. If I may humbly ask, could someone perhaps nudge me in the proper direction? Thanks and have a nice morning. (Also, this is my first time using Github and my programming knowledge and skill is very nascent, so please forgive me if the structuring and explanation of this issue report isn't ideal.)

Steps to reproduce the behavior:

1. Install Whishper and Docker by carefully following the instructions at https://whishper.net/guides/install/.
2. Make sure the docker-compose.yaml file has the local IP address of "http://127.0.0.1:8082".
3. Open the page connected to the aforementioned IP address and try transcribing a video file (whether locally or through a URL).
4. Check for the "transcription failed" error.

Expected behavior

A red box with a phrase indicating that the video transcription failed. (Possible ErrorID: id-6781ffe11d9976e2315349aa)

Environment

• OS: Windows
• Browser: Chrome
• Version: 131.0.6778.265
• Hosting: Whishper's localhost

[docker-compose.txt](https://github.com/user-attachments/files/18384370/docker-compose.txt)

Photo evidence:

To reiterate, the upload is successful, but the transcription isn't.

[JoshuaH-072017]

Let me update this: after changing the IP address in the .env and docker-compose to my local IP address, while transcriptions for audio files now successfully work, it's simply URLs that the program struggles to even try to transcribe, yielding the same "failed to transcribe" error. I know it's not the most important thing to be mildly frustrated over, but could someone please give assistance?

[ShadowLinks]

https://github.com/pluja/whishper/issues/112#issuecomment-2353041194 update yt-dlp in the container.

[JoshuaH-072017]

@ShadowLinks, thanks for helping me with this. I was starting to wonder if anyone would actually notice my post. Well, have a nice evening (or time of day wherever you happen to live).

[stratus-ss]

How do you debug this output? There doesn't seem to be any extra logs

For me it works with CPU but not GPU, although torch seems fine

>>> import torch
>>> print(torch.cuda.is_available())
True
>>> print(torch.cuda.get_device_name(0))
NVIDIA GeForce RTX 5060 Ti
>>> print(torch.version.cuda)
12.8

>>> x = torch.randn(5, 5).cuda()
>>> y = x @ x
>>> print(y)
tensor([[ 4.2216, -2.3547,  6.2181, -5.6701,  0.0156],
        [-0.1115,  2.8345, -0.5160,  0.8715, -0.4545],
        [-0.4388,  2.3835,  1.8378, -8.6422,  3.1802],
        [-0.9043, -0.3692, -1.9726,  3.6068,  1.5088],
        [-1.0358,  4.5116, -6.3428,  6.6190, -0.5088]], device='cuda:0')
>>> print(torch.backends.cuda.matmul.allow_tf32)
False
>>> print(torch.cuda.get_arch_list())
['sm_75', 'sm_80', 'sm_86', 'sm_90', 'sm_100', 'sm_120', 'compute_120']

It'd be nice if there were more logs or output somewhere

[melroy89]

I'm getting "Failed: Could not transcribe."

POST call to http://localhost:8082/api/transcriptions did respond with 200 OK actually.

But the result is null?? and translations is also null..? 😢

I moved to: https://github.com/DevDema/whishper-reloaded