Doubt that. There must be something odd going on with it. My data set didn't work with it, tried two times. Step 11 errors both times, yet running it myself via WSL/Linux working perfectly fine. Considering I'm not the only one who had this issue, doesn't make it a rare outlier. Heck, getting it working in WSL/Linux manually is much harder than using the provided notebook, doubt it's a "skill issue" =P.

I Am Making A Skid And Pump Talknet Model With @IBob012's Datasets, I Am Training Skid First

I Got An Error On Step 9 https://cdn.discordapp.com/attachments/994486437883936809/1101867891562926162/message.txt

i havent used a talknet colab in months i dont think id be able to help

i recommend training on so-vits, youll get a better model and its possible i can even help in case you run into an error

That's funny. I didn't expect to have no quotes on the Talknet section of the actual Uberduck site.

How long were you away from this community?

I don't know.

The TalkNET inference nb has been fixed 👍

How much wavs should a talknet model have? My bluey model has 51 I believe

It's how many wavs. Also, why do people use the number of wav files as a way of measuring voice data? Obviously, the best way to measure voice data is with the total duration of the wav files in a dataset.

For example: On the one hand, you can have 124 wavs in a dataset, but they can total less than 8 minutes in duration. On the other hand, you can have 64 wavs in a dataset, and they can total over 10 minutes in duration.

So yeah, measure wisely.

Oh, and @Gabherelol, my advice is at least 10 minutes of voice data.

Is there a current tutorial on how to train a Talknet model?

don't train a talknet model. keep doing sovits.

Sovits doesn't do my voice justice with certain characters that I use in my series, hence why I wanna try using Talknet as well

ok

What happened to the Beavis and Butt Head models in the reference audio?

The uploader of those models has privated them.

My Sovits 4.0 models are 100% more cleaner and better than those.

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-3-191d0a0af518> in <cell line: 10>()
      8 
      9 import json
---> 10 from nemo.collections.asr.models import EncDecCTCModel
     11 asr_model = EncDecCTCModel.from_pretrained(model_name="asr_talknet_aligner").cpu().eval()
     12 

22 frames
/usr/local/lib/python3.10/dist-packages/google/protobuf/descriptor.py in __new__(cls, name, full_name, index, number, type, cpp_type, label, default_value, message_type, enum_type, containing_type, is_extension, extension_scope, options, serialized_options, has_default_value, containing_oneof, json_name, file, create_key)
    559                 has_default_value=True, containing_oneof=None, json_name=None,
    560                 file=None, create_key=None):  # pylint: disable=redefined-builtin
--> 561       _message.Message._CheckCalledFromGeneratedFile()
    562       if is_extension:
    563         return _message.default_pool.FindExtensionByName(full_name)

TypeError: Descriptors cannot not be created directly.
If this call came from a _pb2.py file, your generated code is out of date and must be regenerated with protoc >= 3.19.0.
If you cannot immediately regenerate your protos, some other possible workarounds are:
 1. Downgrade the protobuf package to 3.20.x or lower.
 2. Set PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python (but this will use pure-Python parsing and will be much slower).

How do I resolve this?

nevermind I fixed it

Oh yeah I forgot to add that line in

Will update in a min

Already added in

can somebody help me with this

Oh @Justin Does the link to the TalkNet training notebook need to be updated in the notebooks channel?

No?

how did you fix it