fixes

requirements.txt

@@ -11,7 +11,6 @@ scikit-learn
 onnxruntime
 pandas
 seaborn
-speechbrain
 transformers
 datasets
 # --------- pytorch --------- #

src/components/losses.py

@@ -6,7 +6,6 @@ import torch.nn.functional as F
 from audiotools import AudioSignal
 from audiotools import STFTParams
 from torch import nn
-from speechbrain.lobes.models.huggingface_transformers.hubert import HuBERT
 
 
 class L1Loss(nn.L1Loss):
@@ -433,25 +432,3 @@ class LossPED(nn.Module):
             "energy_loss": energy_loss * self.energy_loss_weight,
         }
         return loss
-
-
-class HuBERTLoss(nn.Module):
-    """Loss Computation for HuBERT model using MSELoss"""
-
-    def __init__(self, loss_fn : typing.Callable = nn.MSELoss(), device = "cuda"):
-        super().__init__()
-        model_hub = "facebook/hubert-base-ls960"
-        save_path = "pretrained_models/hubert-base-ls960"
-        self.model = HuBERT(model_hub, save_path).to(device)
-        self.loss_fn = loss_fn
-
-    def forward(self, inputs, targets):
-        if isinstance(inputs, AudioSignal):
-            inputs = getattr(inputs, "audio_data")
-            targets = getattr(targets, "audio_data")
-
-        inputs = self.model(inputs.squeeze())
-        targets = self.model(targets.squeeze())
-        
-        loss = self.loss_fn(inputs, targets)
-        return loss