[yesno] Remove padding in TDNN (#21)

* Disable SpecAug for yesno.

Also replace Adam with SGD.

* Remove padding in the model to make the results reproducible.

Author: csukuangfj

.github/workflows/run-yesno-recipe.yml

@@ -69,21 +69,10 @@ jobs:
         run: |
           export PYTHONPATH=$PWD:$PYTHONPATH
           echo $PYTHONPATH
-          ls -lh
 
-          # The following three lines are for macOS
-          lib_path=$(python -c "from distutils.sysconfig import get_python_lib; print(get_python_lib())")
-          echo "lib_path: $lib_path"
-          export DYLD_LIBRARY_PATH=$lib_path:$DYLD_LIBRARY_PATH
-          ls -lh $lib_path
 
           cd egs/yesno/ASR
           ./prepare.sh
-          python3 ./tdnn/train.py --num-epochs 100
-          python3 ./tdnn/decode.py --epoch 99
-          python3 ./tdnn/decode.py --epoch 95
-          python3 ./tdnn/decode.py --epoch 90
-          python3 ./tdnn/decode.py --epoch 80
-          python3 ./tdnn/decode.py --epoch 70
-          python3 ./tdnn/decode.py --epoch 60
+          python3 ./tdnn/train.py
+          python3 ./tdnn/decode.py
           # TODO: Check that the WER is less than some value

egs/yesno/ASR/tdnn/asr_datamodule.py

@@ -27,7 +27,6 @@
     K2SpeechRecognitionDataset,
     PrecomputedFeatures,
     SingleCutSampler,
-    SpecAugment,
 )
 from lhotse.dataset.input_strategies import OnTheFlyFeatures
 from torch.utils.data import DataLoader
@@ -163,18 +162,8 @@ def train_dataloaders(self) -> DataLoader:
                 )
             ] + transforms
 
-        input_transforms = [
-            SpecAugment(
-                num_frame_masks=2,
-                features_mask_size=27,
-                num_feature_masks=2,
-                frames_mask_size=100,
-            )
-        ]
-
         train = K2SpeechRecognitionDataset(
             cut_transforms=transforms,
-            input_transforms=input_transforms,
             return_cuts=self.args.return_cuts,
         )
 
@@ -194,7 +183,6 @@ def train_dataloaders(self) -> DataLoader:
                 input_strategy=OnTheFlyFeatures(
                     Fbank(FbankConfig(num_mel_bins=23))
                 ),
-                input_transforms=input_transforms,
                 return_cuts=self.args.return_cuts,
             )
 

egs/yesno/ASR/tdnn/decode.py

@@ -32,14 +32,14 @@ def get_parser():
     parser.add_argument(
         "--epoch",
         type=int,
-        default=9,
+        default=14,
         help="It specifies the checkpoint to use for decoding."
         "Note: Epoch counts from 0.",
     )
     parser.add_argument(
         "--avg",
         type=int,
-        default=15,
+        default=2,
         help="Number of checkpoints to average. Automatically select "
         "consecutive checkpoints before the checkpoint specified by "
         "'--epoch'. ",
@@ -104,16 +104,11 @@ def decode_one_batch(
     nnet_output = model(feature)
     # nnet_output is [N, T, C]
 
-    supervisions = batch["supervisions"]
-
-    supervision_segments = torch.stack(
-        (
-            supervisions["sequence_idx"],
-            supervisions["start_frame"],
-            supervisions["num_frames"],
-        ),
-        1,
-    ).to(torch.int32)
+    batch_size = nnet_output.shape[0]
+    supervision_segments = torch.tensor(
+        [[i, 0, nnet_output.shape[1]] for i in range(batch_size)],
+        dtype=torch.int32,
+    )
 
     lattice = get_lattice(
         nnet_output=nnet_output,

egs/yesno/ASR/tdnn/model.py

@@ -23,15 +23,13 @@ def __init__(self, num_features: int, num_classes: int):
                 in_channels=num_features,
                 out_channels=32,
                 kernel_size=3,
-                padding=1,
             ),
             nn.ReLU(inplace=True),
             nn.BatchNorm1d(num_features=32, affine=False),
             nn.Conv1d(
                 in_channels=32,
                 out_channels=32,
                 kernel_size=5,
-                padding=4,
                 dilation=2,
             ),
             nn.ReLU(inplace=True),
@@ -40,7 +38,6 @@ def __init__(self, num_features: int, num_classes: int):
                 in_channels=32,
                 out_channels=32,
                 kernel_size=5,
-                padding=8,
                 dilation=4,
             ),
             nn.ReLU(inplace=True),

egs/yesno/ASR/tdnn/train.py

@@ -24,12 +24,7 @@
 from icefall.dist import cleanup_dist, setup_dist
 from icefall.graph_compiler import CtcTrainingGraphCompiler
 from icefall.lexicon import Lexicon
-from icefall.utils import (
-    AttributeDict,
-    encode_supervisions,
-    setup_logger,
-    str2bool,
-)
+from icefall.utils import AttributeDict, setup_logger, str2bool
 
 
 def get_parser():
@@ -61,7 +56,7 @@ def get_parser():
     parser.add_argument(
         "--num-epochs",
         type=int,
-        default=50,
+        default=15,
         help="Number of epochs to train.",
     )
 
@@ -129,11 +124,10 @@ def get_params() -> AttributeDict:
         {
             "exp_dir": Path("tdnn/exp"),
             "lang_dir": Path("data/lang_phone"),
-            "lr": 1e-3,
+            "lr": 1e-2,
             "feature_dim": 23,
             "weight_decay": 1e-6,
             "start_epoch": 0,
-            "num_epochs": 50,
             "best_train_loss": float("inf"),
             "best_valid_loss": float("inf"),
             "best_train_epoch": -1,
@@ -278,9 +272,14 @@ def compute_loss(
     # different duration in decreasing order, required by
     # `k2.intersect_dense` called in `k2.ctc_loss`
     supervisions = batch["supervisions"]
-    supervision_segments, texts = encode_supervisions(
-        supervisions, subsampling_factor=1
+    texts = supervisions["text"]
+
+    batch_size = nnet_output.shape[0]
+    supervision_segments = torch.tensor(
+        [[i, 0, nnet_output.shape[1]] for i in range(batch_size)],
+        dtype=torch.int32,
     )
+
     decoding_graph = graph_compiler.compile(texts)
 
     dense_fsa_vec = k2.DenseFsaVec(
@@ -491,7 +490,7 @@ def run(rank, world_size, args):
     if world_size > 1:
         model = DDP(model, device_ids=[rank])
 
-    optimizer = optim.AdamW(
+    optimizer = optim.SGD(
         model.parameters(),
         lr=params.lr,
         weight_decay=params.weight_decay,