lol yea a single 2 hour long wav

I don’t think that’s how it works

yea i dont know how things work

Must be split up between clips of various lengths.

Follow the tutorial, then start going into the notebooks

I don't blame you. There aren't any tutorials linked in plain sight on uberduck.ai.

i remember looking at uberduck's website for the tutorial but it was outdated hence im here

maybe there is a tutorial link here on discord that i havent noticed yet?

You mean like the one that's linked in #841422801965416538?

what might be the issue for this error? i dont think its due to incorrect sampling rate. do i need to decompress the wav files? its error from transcribing notebook

Error                                     Traceback (most recent call last)
/usr/local/lib/python3.9/dist-packages/whisper/audio.py in load_audio(file, sr)
     41         out, _ = (
---> 42             ffmpeg.input(file, threads=0)
     43             .output("-", format="s16le", acodec="pcm_s16le", ac=1, ar=sr)

4 frames
Error: ffmpeg error (see stderr output for detail)

During handling of the above exception, another exception occurred:

RuntimeError                              Traceback (most recent call last)
/usr/local/lib/python3.9/dist-packages/whisper/audio.py in load_audio(file, sr)
     45         )
     46     except ffmpeg.Error as e:
---> 47         raise RuntimeError(f"Failed to load audio: {e.stderr.decode()}")
     48 
     49     return np.frombuffer(out, np.int16).flatten().astype(np.float32) / 32768.0

RuntimeError: Failed to load audio: ffmpeg version 4.2.7-0ubuntu0.1 Copyright (c) 2000-2022 the 
wavs/wavs.zip: Invalid data found when processing input

First, are you using one big file or cut pieces of audio to transcribe?

just a bunch of 19 seconds wav files

roughly 400 files

inside the zip

You will need 15 seconds or less in each file

otherwise you won't be able to train

swag

what is the limit on how many files free version of google colab can handle?

should these files be named in a specific way

No. It's easier to use number labels like

1.wav

,

2.wav

, etc

So most people do that

i should upload the file as zip right? i shouldnt decompress all the wav files inside

The whole dataset should be just 1 .zip

I don't think it expects nested zips?

dont have any nested zips or folders within, just wav files

12*

That's the cutoff duration

T4 can handle 15 seconds

Oh

If the batch size is 18 or lower