add more aug

amt/audio.py

@@ -188,7 +188,7 @@ def __init__(
         reverb_factor: int = 1,
         min_snr: int = 10,
         max_snr: int = 40,
-        max_pitch_shift: int = 4,
+        max_pitch_shift: int = 5,
     ):
         super().__init__()
         self.tokenizer = AmtTokenizer()
@@ -232,10 +232,10 @@ def __init__(
         # Spec aug
         self.spec_aug = torch.nn.Sequential(
             torchaudio.transforms.FrequencyMasking(
-                freq_mask_param=10, iid_masks=True
+                freq_mask_param=15, iid_masks=True
             ),
             torchaudio.transforms.TimeMasking(
-                time_mask_param=100, iid_masks=True
+                time_mask_param=500, iid_masks=True
             ),
         )
 

amt/train.py

@@ -339,7 +339,8 @@ def train_loop(
 
             wav, src, tgt = batch  # (b_sz, s_len), (b_sz, s_len, v_sz)
             with torch.no_grad():
-                mel, (src, tgt) = audio_transform.forward(wav, src, tgt)
+                mel = audio_transform.mel(wav)
+                # mel, (src, tgt) = audio_transform.forward(wav, src, tgt)
             logits = model(mel, src)  # (b_sz, s_len, v_sz)
             logits = logits.transpose(1, 2)  # Transpose for CrossEntropyLoss
             loss = loss_fn(logits, tgt)

config/config.json

@@ -17,7 +17,7 @@
         "n_mels": 256
     },
     "data": {
-        "stride_factor": 1,
+        "stride_factor": 3,
         "max_seq_len": 4096
     }
 }