Yae Miko Encanto

I don't know who made this model, but I have to admit that this weird bass in the background adds an interesting and dark touch 😳

anyone know if it's possible to run tacotron training and then later do talknet training on the same checkpoint?

Tacotron 2 and TalkNet are different voice model architectures.

dang, ok

If i dont hear any robotic static sounds that means is good

Yes Tacotron also makes expressions if you have the dataset

For the RunTimeError I keep getting. When you say upload text file, you mean the transcription list right? As for custom inferencing config, would you say I'm doing this part wrong? # This is your training config. # Notes: # Change n_speakers if running multispeaker. # Change training_audiopaths_and_text and val_audiopaths_and_text to proper transcript file name. # If you are training a multispeaker model: # Add your sample_inference_speaker_ids in a list, e.g: [0, 1, 2, 3] # And make sure your metadata file (transcript) has the appropriate |0 |1 |2 |3 suffixes! %%writefile tacotron2_config.json { "batch_size": 18, "checkpoint_name": "uberduck_model", "checkpoint_path": "/content/checkpoints", "cudnn_enabled": true, "dataset_path": ".", "debug": false, "distributed_run": false, "epochs": 100, "epochs_per_checkpoint": 20, "fp16_run": false, "include_f0": false, "learning_rate": 5e-4, "log_dir": "runs", "n_speakers": 1, "p_arpabet": 1.0, "has_speaker_embedding": true, "sample_inference_speaker_ids": [0], "sample_rate": 22050, "steps_per_sample": 50, "text_cleaners": ["english_cleaners"], "max_decoder_steps": 1000, "training_audiopaths_and_text": "transcription.txt", "val_audiopaths_and_text": "transcription.txt", "warm_start_name": "tacotron2_statedict.pt", "ignore_layers": ["speaker_embedding.weight"], "seed": 123, "gst_dim": 2304, "gst_type": "torchmoji", "torchmoji_vocabulary_file":"vocabulary.json", "torchmoji_model_file": "pytorch_model.bin" }

i don't mean the transcription list, i mean the config file

select everything between {these brackets} (including those brackets), copy that, paste it into a new file

then save the file, and when it asks you for config then upload it

thank you! I think that got me another step closer. I now got this new error. Any ideas what it still needs?

iiiinteresting

uhh set

ignore_layers

to

["null"]

in the config for one

omg it works now! THANK YOU

oh dang

What are the dialogue I should add to Expression sounds like Gasp Screaming aaaaaaaa Grunts Crying

You COULD get screaming by just writing AAAA's or "Ah!" but other then that I wouldn't transcribe gasping, grunting or crying

@Pikachu ✓ why your son so rude 😔...

That's not my son

you heard it here first

pikachu has disowned his son pichu

how many voice clips are needed for a good model, and how long should they be.

around 100-200 or more.

5-12.

5-12 seconds each

cringe

stop the fnf gorebage