Changed rounding to 0 to avoid AssertionError on line 567

[NickAnastasoff]

No description provided.

[dagshub]

Join the discussion on DagsHub!

[Tywors]

Same issue:

AssertionError Traceback (most recent call last)
in <cell line: 2>()
1 # simple generation
----> 2 audio_array = generate_audio(text_prompt, history_prompt=voice_name, text_temp=0.7, waveform_temp=0.7)

2 frames
/content/bark-with-voice-clone/bark/generation.py in generate_coarse(x_semantic, history_prompt, temp, top_k, top_p, silent, max_coarse_history, sliding_window_len, use_kv_caching)
565 and x_coarse_history.max() <= CODEBOOK_SIZE - 1
566 and (
--> 567 round(x_coarse_history.shape[-1] / len(x_semantic_history), 0)
568 == round(semantic_to_coarse_ratio / N_COARSE_CODEBOOKS, 0)
569 )

AssertionError:

[francislabountyjr]

This may just be coming from the fact that the semantic token to coarse token ratio is not correct. We are working on getting HuBert working with bark so we don't have to use a hacky solution for generating the semantic tokens. Will also open up finetuning once it's done as well. Not a very hard or time consuming project, just been busy with other things. Will hopefully have that out in a couple days or a week or so unless someone else completes that first (someone is working on it here https://github.com/EndlessReform/bark-with-voice-clone/tree/semantic_codebook)