Commit 0b078214 authored by Phil Wang's avatar Phil Wang
Browse files

use a hack to fix an issue, refactor semantic later

parent 9c4c56d3

README.md

+1 −0
Original line number Diff line number Diff line
@@ -215,6 +215,7 @@ generated_wav_with_text_condition = audiolm(text = ['chirping of birds and the d 
- [ ] add option to use flash attention

- [ ] simplify training even more within AudioLM class

- [ ] cli tool, something like `audiolm generate <wav.file | text>` and save generated wav file to local directory

- [ ] refactor so semantic transformer has a wrapper to that handles unique consecutives as well as wav to hubert or vq-wav2vec



## Citations

audiolm_pytorch/audiolm_pytorch.py

+5 −2
Original line number Diff line number Diff line
@@ -410,6 +410,7 @@ class SemanticTransformer(nn.Module): 
            logits = self.forward_with_cond_scale(

                ids = sample_semantic_ids,

                text_embeds = text_embeds,

                unique_consecutive = False,

                **kwargs

            )
@@ -467,9 +468,11 @@ class SemanticTransformer(nn.Module): 
        return_loss = False,

        text: Optional[List[str]] = None,

        text_embeds = None,

        cond_drop_prob = None

        cond_drop_prob = None,

        unique_consecutive = None

    ):

        device = self.device

        unique_consecutive = default(unique_consecutive, self.unique_consecutive)



        assert exists(raw_wave) ^ exists(ids)
@@ -482,7 +485,7 @@ class SemanticTransformer(nn.Module): 
        if self.training:

            ids = append_eos_id(ids, self.eos_id)



        if self.unique_consecutive:

        if unique_consecutive:

            ids = batch_unique_consecutive(ids, pad_value = self.pad_id)



        has_text = exists(text) or exists(text_embeds)

setup.py

+1 −1
Original line number Diff line number Diff line
@@ -3,7 +3,7 @@ from setuptools import setup, find_packages 
setup(

  name = 'audiolm-pytorch',

  packages = find_packages(exclude=[]),

  version = '0.0.42',

  version = '0.0.43',

  license='MIT',

  description = 'AudioLM - Language Modeling Approach to Audio Generation from Google Research - Pytorch',

  author = 'Phil Wang',