Some gui parameters not passed properly

[ThrowawayAccount01]

From version 1.3.3 onwards, using old trained models will result in audio distortions.

I use the following settings:

[14:17:52] INFO     [14:17:52] Event infer, values {'model_path': 'C:/Users/User/Desktop/hapi/G_21600.pth',   gui.py:509
                    'model_path_browse': 'C:/Users/User/Desktop/hapi/G_21600.pth', 'config_path':
                    'C:/Users/User/Desktop/hapi/config.json', 'config_path_browse':
                    'C:/Users/User/Desktop/hapi/config.json', 'cluster_model_path':
                    'C:/Users/User/Desktop/hapi/kmeans.pt', 'cluster_model_path_browse':
                    'C:/Users/User/Desktop/hapi/kmeans.pt', 'speaker': 'hapiraw', 'silence_threshold': -35.0,
                    'transpose': 8.0, 'auto_predict_f0': False, 'f0_method': 'crepe', 'cluster_infer_ratio':
                    0.5, 'noise_scale': 0.4, 'pad_seconds': 0.1, 'chunk_seconds': 0.5, 'absolute_thresh':
                    True, 'input_path': 'C:/Users/User/Desktop/oblivion/kai.wav', 'input_path_browse':
                    'C:/Users/User/Desktop/oblivion/kai.wav', 'auto_play': False, 'crossfade_seconds': 0.05,
                    'block_seconds': 0.35, 'additional_infer_before_seconds': 0.15,
                    'additional_infer_after_seconds': 0.1, 'realtime_algorithm': '1 (Divide constantly)',
                    'input_device': 'Microsoft Sound Mapper - Input (MME)', 'output_device': 'SteelSeries
                    Sonar - Media (Stee (MME)', 'passthrough_original': False, 'presets': 'Default VC (GPU,
                    GTX 1060)', 'preset_name': '', 'use_gpu': True}
[14:17:53] INFO     [14:17:53] current directory is C:\Users\User\Desktop\sovits\venv\Scripts  hubert_pretraining.py:116
           INFO     [14:17:53] HubertPretrainingTask Config {'_name': 'hubert_pretraining',    hubert_pretraining.py:117
                    'data': 'metadata', 'fine_tuning': False, 'labels': ['km'], 'label_dir':
                    'label', 'label_rate': 50.0, 'sample_rate': 16000, 'normalize': False,
                    'enable_padding': False, 'max_keep_size': None, 'max_sample_size': 250000,
                    'min_sample_size': 32000, 'single_target': False, 'random_crop': True,
                    'pad_audio': False}
           INFO     [14:17:53] HubertModel Config: {'_name': 'hubert', 'label_rate': 50.0,                 hubert.py:250
                    'extractor_mode': default, 'encoder_layers': 12, 'encoder_embed_dim': 768,
                    'encoder_ffn_embed_dim': 3072, 'encoder_attention_heads': 12, 'activation_fn': gelu,
                    'layer_type': transformer, 'dropout': 0.1, 'attention_dropout': 0.1,
                    'activation_dropout': 0.0, 'encoder_layerdrop': 0.05, 'dropout_input': 0.1,
                    'dropout_features': 0.1, 'final_dim': 256, 'untie_final_proj': True,
                    'layer_norm_first': False, 'conv_feature_layers': '[(512,10,5)] + [(512,3,2)] * 4 +
                    [(512,2,2)] * 2', 'conv_bias': False, 'logit_temp': 0.1, 'target_glu': False,
                    'feature_grad_mult': 0.1, 'mask_length': 10, 'mask_prob': 0.8, 'mask_selection':
                    static, 'mask_other': 0.0, 'no_mask_overlap': False, 'mask_min_space': 1,
                    'mask_channel_length': 10, 'mask_channel_prob': 0.0, 'mask_channel_selection': static,
                    'mask_channel_other': 0.0, 'no_mask_channel_overlap': False, 'mask_channel_min_space':
                    1, 'conv_pos': 128, 'conv_pos_groups': 16, 'latent_temp': [2.0, 0.5, 0.999995],
                    'skip_masked': False, 'skip_nomask': False, 'checkpoint_activations': False,
                    'required_seq_len_multiple': 2, 'depthwise_conv_kernel_size': 31, 'attn_type': '',
                    'pos_enc_type': 'abs', 'fp16': False}

I believe it has something to do with the commit that changed the defaults to dio?

Here are some samples to show the distortions. I used the exact same settings for both inferences:

1.3.5: https://voca.ro/1eR1lyRN585P

1.3.3: https://voca.ro/1kniA71h8KMa

As you can hear, 1.3.5 has some weird pitch shifts and vocals.

[34j]

Reverted 1.3.5 changes. Is it fixed?

[ThrowawayAccount01]

Updated to 1.3.6, the result sounds exactly the same as 1.3.5. Still has the same weird audio distortions.

This problem started happening from 1.3.4, when the defaults were changed to dio.

[34j]

So I have to revert revert commits...
What about using dio in 1.3.3? Isn't it simply a matter of dio performance?

[ThrowawayAccount01]

In 1.3.3, setting to either crepe or dio works fine. The sample I posted uses crepe, but when using dio it sounds similar.

From 1.3.4 and up, using either crepe or dio will produce the weird distortions that you can hear from my 1.3.5 sample.

[34j]

I found that this is a bug where a parameter in the GUI was not properly passed. The distortion is a dio issue. I will fix it soon.

[34j]

Well found as I think every time you submit a issue.