add decoding

egs/spgispeech/ASR/conformer_ctc/asr_datamodule.py

@@ -104,7 +104,7 @@ class SPGISpeechAsrDataModule:
         group.add_argument(
             "--max-duration",
             type=int,
-            default=140.0,
+            default=100.0,
             help="Maximum pooled recordings duration (seconds) in a "
             "single batch. You can reduce it if it causes CUDA OOM.",
         )
@@ -296,7 +296,7 @@ class SPGISpeechAsrDataModule:
         return load_manifest_lazy(self.args.manifest_dir / "cuts_dev.jsonl.gz")
 
     @lru_cache()
-    def test_cuts(self) -> CutSet:
+    def val_cuts(self) -> CutSet:
         logging.info("About to get SPGISpeech val cuts")
         return load_manifest_lazy(self.args.manifest_dir / "cuts_val.jsonl.gz")
 

egs/spgispeech/ASR/conformer_ctc/decode.py

@@ -23,10 +23,11 @@ from pathlib import Path
 from typing import Dict, List, Optional, Tuple
 
 import k2
+from numpy import True_
 import sentencepiece as spm
 import torch
 import torch.nn as nn
-from asr_datamodule import LibriSpeechAsrDataModule
+from asr_datamodule import SPGISpeechAsrDataModule
 from conformer import Conformer
 
 from icefall.bpe_graph_compiler import BpeCtcTrainingGraphCompiler
@@ -131,7 +132,7 @@ def get_parser():
     parser.add_argument(
         "--lang-dir",
         type=str,
-        default="data/lang_bpe_500",
+        default="data/lang_bpe_5000",
         help="The lang dir",
     )
 
@@ -140,7 +141,7 @@ def get_parser():
         type=str,
         default="data/lm",
         help="""The LM dir.
-        It should contain either G_4_gram.pt or G_4_gram.fst.txt
+        It should contain either G_3_gram.pt or G_3_gram.fst.txt
         """,
     )
 
@@ -420,7 +421,7 @@ def decode_dataset(
       G:
         An LM. It is not None when params.method is "nbest-rescoring"
         or "whole-lattice-rescoring". In general, the G in HLG
-        is a 3-gram LM, while this G is a 4-gram LM.
+        is a 3-gram LM.
     Returns:
       Return a dict, whose key may be "no-rescore" if no LM rescoring
       is used, or it may be "lm_scale_0.7" if LM rescoring is used.
@@ -462,9 +463,7 @@ def decode_dataset(
 
                 results[lm_scale].extend(this_batch)
         else:
-            assert (
-                len(results) > 0
-            ), "It should not decode to empty in the first batch!"
+            assert len(results) > 0, "It should not decode to empty in the first batch!"
             this_batch = []
             hyp_words = []
             for ref_text in texts:
@@ -479,9 +478,7 @@ def decode_dataset(
         if batch_idx % 100 == 0:
             batch_str = f"{batch_idx}/{num_batches}"
 
-            logging.info(
-                f"batch {batch_str}, cuts processed until now is {num_cuts}"
-            )
+            logging.info(f"batch {batch_str}, cuts processed until now is {num_cuts}")
     return results
 
 
@@ -512,9 +509,7 @@ def save_results(
             test_set_wers[key] = wer
 
         if enable_log:
-            logging.info(
-                "Wrote detailed error stats to {}".format(errs_filename)
-            )
+            logging.info("Wrote detailed error stats to {}".format(errs_filename))
 
     test_set_wers = sorted(test_set_wers.items(), key=lambda x: x[1])
     errs_info = params.exp_dir / f"wer-summary-{test_set_name}.txt"
@@ -534,7 +529,7 @@ def save_results(
 @torch.no_grad()
 def main():
     parser = get_parser()
-    LibriSpeechAsrDataModule.add_arguments(parser)
+    SPGISpeechAsrDataModule.add_arguments(parser)
     args = parser.parse_args()
     args.exp_dir = Path(args.exp_dir)
     args.lang_dir = Path(args.lang_dir)
@@ -570,7 +565,7 @@ def main():
         HLG = None
         H = k2.ctc_topo(
             max_token=max_token_id,
-            modified=False,
+            modified=True,  # Use modified topology since vocab size is large
             device=device,
         )
         bpe_model = spm.SentencePieceProcessor()
@@ -591,10 +586,10 @@ def main():
         "whole-lattice-rescoring",
         "attention-decoder",
     ):
-        if not (params.lm_dir / "G_4_gram.pt").is_file():
-            logging.info("Loading G_4_gram.fst.txt")
+        if not (params.lm_dir / "G_3_gram.pt").is_file():
+            logging.info("Loading G_3_gram.fst.txt")
             logging.warning("It may take 8 minutes.")
-            with open(params.lm_dir / "G_4_gram.fst.txt") as f:
+            with open(params.lm_dir / "G_3_gram.fst.txt") as f:
                 first_word_disambig_id = lexicon.word_table["#0"]
 
                 G = k2.Fsa.from_openfst(f.read(), acceptor=False)
@@ -615,10 +610,10 @@ def main():
                 # for why we need to do this.
                 G.dummy = 1
 
-                torch.save(G.as_dict(), params.lm_dir / "G_4_gram.pt")
+                torch.save(G.as_dict(), params.lm_dir / "G_3_gram.pt")
         else:
-            logging.info("Loading pre-compiled G_4_gram.pt")
-            d = torch.load(params.lm_dir / "G_4_gram.pt", map_location=device)
+            logging.info("Loading pre-compiled G_3_gram.pt")
+            d = torch.load(params.lm_dir / "G_3_gram.pt", map_location=device)
             G = k2.Fsa.from_dict(d)
 
         if params.method in ["whole-lattice-rescoring", "attention-decoder"]:
@@ -662,16 +657,16 @@ def main():
     num_param = sum([p.numel() for p in model.parameters()])
     logging.info(f"Number of model parameters: {num_param}")
 
-    librispeech = LibriSpeechAsrDataModule(args)
+    spgispeech = SPGISpeechAsrDataModule(args)
 
-    test_clean_cuts = librispeech.test_clean_cuts()
-    test_other_cuts = librispeech.test_other_cuts()
+    dev_cuts = spgispeech.dev_cuts()
+    val_cuts = spgispeech.val_cuts()
 
-    test_clean_dl = librispeech.test_dataloaders(test_clean_cuts)
-    test_other_dl = librispeech.test_dataloaders(test_other_cuts)
+    dev_dl = spgispeech.test_dataloaders(dev_cuts)
+    val_dl = spgispeech.test_dataloaders(val_cuts)
 
-    test_sets = ["test-clean", "test-other"]
-    test_dl = [test_clean_dl, test_other_dl]
+    test_sets = ["dev", "val"]
+    test_dl = [dev_dl, val_dl]
 
     for test_set, test_dl in zip(test_sets, test_dl):
         results_dict = decode_dataset(
@@ -687,9 +682,7 @@ def main():
             eos_id=eos_id,
         )
 
-        save_results(
-            params=params, test_set_name=test_set, results_dict=results_dict
-        )
+        save_results(params=params, test_set_name=test_set, results_dict=results_dict)
 
     logging.info("Done!")
 

egs/spgispeech/ASR/decode.sh

@@ -0,0 +1,10 @@
+#!/usr/bin/env bash
+
+set -eou pipefail
+
+. ./path.sh
+. parse_options.sh || exit 1
+
+# Train Conformer CTC model
+utils/queue-freegpu.pl --gpu 1 --mem 10G -l "hostname=c*" -q g.q conformer_ctc/exp/decode.log \
+  python conformer_ctc/decode.py --epoch 12 --avg 3 --method ctc-decoding --max-duration 50 --num-paths 20