First working example

egs/librispeech/ASR/zipformer/asr_datamodule_with_parallel_aug.py

@@ -28,7 +28,6 @@ import torch
 from lhotse import CutSet, Fbank, FbankConfig, load_manifest_lazy
 from lhotse.cut import Cut
 from lhotse.dataset import (  # noqa F401 for PrecomputedFeatures
-    CutConcatenate,
     CutMix,
     DynamicBucketingSampler,
     K2SpeechRecognitionDataset,
@@ -57,21 +56,33 @@ class _SeedWorkers:
         fix_random_seed(self.seed + worker_id)
 
 
+"""
+We use c.features = None below to suppress the following warnings
+
+2025-05-29 16:49:55,253 WARNING [data.py:801] Attempting to perturb speed on a
+DataCut that references pre-computed features. The feature manifest will be
+detached, as we do not support feature-domain speed perturbation.
+"""
+
+
 def perturb_speed(c: Cut):
-    factor = random.uniform(0.9, 1.1)
-    print("perturb_speed factor", factor)
+    factor = random.choice([0.9, 1.1])
+    c.features = None
+
     return lhotse.MonoCut.perturb_speed(c, factor)
 
 
 def perturb_volume(c: Cut):
-    factor = random.uniform(0.9, 1.1)
-    print("perturb_volume factor", factor)
+    factor = random.choice([0.9, 1.1])
+    c.features = None
+
     return lhotse.MonoCut.perturb_volume(c, factor)
 
 
 def perturb_tempo(c: Cut):
-    factor = random.uniform(0.9, 1.1)
-    print("perturb_tempo factor", factor)
+    factor = random.choice([0.9, 1.1])
+
+    c.features = None
     return lhotse.MonoCut.perturb_tempo(c, factor)
 
 
@@ -86,7 +97,6 @@ class LibriSpeechAsrDataModuleWithParallelAug:
     experiments, e.g.:
     - dynamic batch size,
     - bucketing samplers,
-    - cut concatenation,
     - augmentation,
     - on-the-fly feature extraction
 
@@ -112,6 +122,12 @@ class LibriSpeechAsrDataModuleWithParallelAug:
             help="""Used only when --mini-libri is False.When enabled,
             use 960h LibriSpeech. Otherwise, use 100h subset.""",
         )
+        group.add_argument(
+            "--enable-augmentation",
+            type=str2bool,
+            default=True,
+            help="True to enable augmentation for training set",
+        )
         group.add_argument(
             "--mini-libri",
             type=str2bool,
@@ -204,7 +220,12 @@ class LibriSpeechAsrDataModuleWithParallelAug:
           sampler_state_dict:
             The state dict for the training sampler.
         """
-        transforms = [perturb_speed, perturb_volume, perturb_tempo]
+        if self.args.enable_augmentation:
+            logging.info("Augmentation is enabled")
+            transforms = [perturb_speed, perturb_volume, perturb_tempo]
+        else:
+            logging.info("Augmentation is disabled")
+            transforms = []
 
         logging.info("About to create train dataset")
         train = ConsistencyRegularizationSpeechRecognitionDataset(
@@ -254,12 +275,6 @@ class LibriSpeechAsrDataModuleWithParallelAug:
 
     def valid_dataloaders(self, cuts_valid: CutSet) -> DataLoader:
         transforms = []
-        if self.args.concatenate_cuts:
-            transforms = [
-                CutConcatenate(
-                    duration_factor=self.args.duration_factor, gap=self.args.gap
-                )
-            ] + transforms
 
         logging.info("About to create dev dataset")
         if self.args.on_the_fly_feats:

egs/librispeech/ASR/zipformer/train_with_aug.py

@@ -855,19 +855,38 @@ def compute_loss(
         disables autograd.
     """
     device = model.device if isinstance(model, DDP) else next(model.parameters()).device
-    feature = batch["inputs"]
+
+    feature_len_seq = [batch["supervisions"]["num_frames"]]
+    text_seq = list(batch["supervisions"]["text"])
+    feature_seq = torch.nn.utils.rnn.unpad_sequence(
+        batch["inputs"],
+        batch["supervisions"]["num_frames"],
+        batch_first=True,
+    )
+
+    if "aug" in batch:
+        for aug in batch["aug"]:
+            feature_len_seq.append(aug["supervisions"]["num_frames"])
+            text_seq.extend(aug["supervisions"]["text"])
+
+            feature_seq.extend(
+                torch.nn.utils.rnn.unpad_sequence(
+                    aug["inputs"],
+                    aug["supervisions"]["num_frames"],
+                    batch_first=True,
+                )
+            )
+
+    feature_lens = torch.cat(feature_len_seq).to(device)
+    feature = torch.nn.utils.rnn.pad_sequence(feature_seq, batch_first=True).to(device)
+
     # at entry, feature is (N, T, C)
     assert feature.ndim == 3
-    feature = feature.to(device)
-
-    supervisions = batch["supervisions"]
-    feature_lens = supervisions["num_frames"].to(device)
 
     batch_idx_train = params.batch_idx_train
     warm_step = params.warm_step
 
-    texts = batch["supervisions"]["text"]
-    y = sp.encode(texts, out_type=int)
+    y = sp.encode(text_seq, out_type=int)
     y = k2.RaggedTensor(y)
 
     with torch.set_grad_enabled(is_training):
@@ -1029,6 +1048,9 @@ def train_one_epoch(
         params.batch_idx_train += 1
         batch_size = len(batch["supervisions"]["text"])
 
+        if "aug" in batch:
+            batch_size *= len(batch["aug"]) + 1
+
         try:
             with torch.cuda.amp.autocast(
                 enabled=params.use_autocast, dtype=params.dtype
@@ -1360,7 +1382,7 @@ def run(rank, world_size, args):
     valid_cuts += librispeech.dev_other_cuts()
     valid_dl = librispeech.valid_dataloaders(valid_cuts)
 
-    if not params.print_diagnostics:
+    if False and not params.print_diagnostics:
         scan_pessimistic_batches_for_oom(
             model=model,
             train_dl=train_dl,
@@ -1443,12 +1465,32 @@ def display_and_save_batch(
     logging.info(f"Saving batch to {filename}")
     torch.save(batch, filename)
 
-    supervisions = batch["supervisions"]
-    features = batch["inputs"]
+    feature_len_seq = [batch["supervisions"]["num_frames"]]
+    text_seq = list(batch["supervisions"]["text"])
+    feature_seq = torch.nn.utils.rnn.unpad_sequence(
+        batch["inputs"],
+        batch["supervisions"]["num_frames"],
+        batch_first=True,
+    )
+
+    if "aug" in batch:
+        for aug in batch["aug"]:
+            feature_len_seq.append(aug["supervisions"]["num_frames"])
+            text_seq.extend(aug["supervisions"]["text"])
+
+            feature_seq.extend(
+                torch.nn.utils.rnn.unpad_sequence(
+                    aug["inputs"],
+                    aug["supervisions"]["num_frames"],
+                    batch_first=True,
+                )
+            )
+
+    features = torch.nn.utils.rnn.pad_sequence(feature_seq, batch_first=True)
 
     logging.info(f"features shape: {features.shape}")
 
-    y = sp.encode(supervisions["text"], out_type=int)
+    y = sp.encode(text_seq, out_type=int)
     num_tokens = sum(len(i) for i in y)
     logging.info(f"num tokens: {num_tokens}")