Add whole-lattice rescoring.

egs/librispeech/ASR/conformer_ctc/decode.py

@@ -42,6 +42,7 @@ from icefall.decode2 import (
     nbest_decoding,
     nbest_oracle as nbest_oracle2,
     rescore_with_n_best_list as rescore_with_n_best_list2,
+    rescore_with_whole_lattice as rescore_with_whole_lattice2,
 )
 from icefall.lexicon import Lexicon
 from icefall.utils import (
@@ -261,9 +262,7 @@ def decode_one_batch(
             )
             hyps = get_texts(best_path)
             hyps = [[word_table[i] for i in ids] for ids in hyps]
-            key = (
+            key = f"oracle_{num_paths}_lattice_score_scale_{params.lattice_score_scale}"  # noqa
-                f"oracle_{num_paths}_lattice_score_scale_{lattice_score_scale}"
-            )
             return {key: hyps}
         else:
             return nbest_oracle(
@@ -322,8 +321,18 @@ def decode_one_batch(
                 scale=params.lattice_score_scale,
             )
     elif params.method == "whole-lattice-rescoring":
+        if True:
+            # TODO: remove "else" branch
+            best_path_dict = rescore_with_whole_lattice2(
+                lattice=lattice,
+                G_with_epsilon_loops=G,
+                lm_scale_list=lm_scale_list,
+            )
+        else:
             best_path_dict = rescore_with_whole_lattice(
-            lattice=lattice, G_with_epsilon_loops=G, lm_scale_list=lm_scale_list
+                lattice=lattice,
+                G_with_epsilon_loops=G,
+                lm_scale_list=lm_scale_list,
             )
     elif params.method == "attention-decoder":
         # lattice uses a 3-gram Lm. We rescore it with a 4-gram LM.
@@ -345,10 +354,14 @@ def decode_one_batch(
         assert False, f"Unsupported decoding method: {params.method}"
 
     ans = dict()
+    if best_path_dict is not None:
         for lm_scale_str, best_path in best_path_dict.items():
             hyps = get_texts(best_path)
             hyps = [[word_table[i] for i in ids] for ids in hyps]
             ans[lm_scale_str] = hyps
+    else:
+        for lm_scale in lm_scale_list:
+            ans[lm_scale_str] = [[] * lattice.shape[0]]
     return ans
 
 

icefall/decode2.py

@@ -17,7 +17,8 @@
 # NOTE: This file is a refactor of decode.py
 # We will delete decode.py and rename this file to decode.py
 
-from typing import Dict, List
+import logging
+from typing import Dict, List, Optional, Union
 
 import k2
 import torch
@@ -505,6 +506,27 @@ def rescore_with_n_best_list(
 ) -> Dict[str, k2.Fsa]:
     """Rescore a nbest list with an n-gram LM.
     The path with a maximum score is used as the decoding output.
+
+    Args:
+      lattice:
+        An FsaVec with axes [utt][state][arc]. It must have the following
+        attributes: ``aux_labels`` and ``lm_scores``. Its labels are
+        token IDs and ``aux_labels`` word IDs.
+      G:
+        An FsaVec containing only a single FSA. It is an n-gram LM.
+      num_paths:
+        Size of nbest list.
+      lm_scale_list:
+        A list of float representing LM score scales.
+      lattice_score_scale:
+        Scale to be applied to ``lattice.score`` when sampling paths
+        using ``k2.random_paths``.
+      use_double_scores:
+        True to use double precision during computation. False to use
+        single precision.
+    Returns:
+      A dict of FsaVec, whose key is an lm_scale and the value is the
+      best decoding path for each utterance in the lattice.
     """
     device = lattice.device
 
@@ -543,3 +565,77 @@ def rescore_with_n_best_list(
         key = f"lm_scale_{lm_scale}"
         ans[key] = best_path
     return ans
+
+
+def rescore_with_whole_lattice(
+    lattice: k2.Fsa,
+    G_with_epsilon_loops: k2.Fsa,
+    lm_scale_list: Optional[List[float]] = None,
+    use_double_scores: bool = True,
+) -> Union[k2.Fsa, Dict[str, k2.Fsa]]:
+    # This is not an Nbest based coding method
+    assert hasattr(lattice, "lm_scores")
+    assert G_with_epsilon_loops.shape == (1, None, None)
+
+    device = lattice.device
+    lattice.scores = lattice.scores - lattice.lm_scores
+    # We will use lm_scores from G, so remove lats.lm_scores here
+    del lattice.lm_scores
+
+    assert hasattr(G_with_epsilon_loops, "lm_scores")
+
+    # Now, lattice.scores contains only am_scores
+
+    # inv_lattice has word IDs as labels.
+    # Its aux_labels are token IDs
+    inv_lattice = k2.invert(lattice)
+    num_seqs = lattice.shape[0]
+
+    b_to_a_map = torch.zeros(num_seqs, device=device, dtype=torch.int32)
+
+    max_loop_count = 10
+    loop_count = 0
+    while loop_count <= max_loop_count:
+        loop_count += 1
+        try:
+            rescoring_lattice = k2.intersect_device(
+                G_with_epsilon_loops,
+                inv_lattice,
+                b_to_a_map,
+                sorted_match_a=True,
+            )
+            rescoring_lattice = k2.top_sort(k2.connect(rescoring_lattice))
+            break
+        except RuntimeError as e:
+            logging.info(f"Caught exception:\n{e}\n")
+            logging.info(
+                f"num_arcs before pruning: {inv_lattice.arcs.num_elements()}"
+            )
+
+            # NOTE(fangjun): The choice of the threshold 1e-9 is arbitrary here
+            # to avoid OOM. You may need to fine tune it.
+            inv_lattice = k2.prune_on_arc_post(inv_lattice, 1e-9, True)
+            logging.info(
+                f"num_arcs after pruning: {inv_lattice.arcs.num_elements()}"
+            )
+    if loop_count > max_loop_count:
+        logging.info("Return None as the resulting lattice is too large")
+        return None
+
+    # lat has token IDs as labels
+    # and word IDs as aux_labels.
+    lat = k2.invert(rescoring_lattice)
+
+    if lm_scale_list is None:
+        return lat
+
+    ans = dict()
+    saved_am_scores = lat.scores - lat.lm_scores
+    for lm_scale in lm_scale_list:
+        am_scores = saved_am_scores / lm_scale
+        lat.scores = am_scores + lat.lm_scores
+
+        best_path = k2.shortest_path(lat, use_double_scores=use_double_scores)
+        key = f"lm_scale_{lm_scale}_yy"
+        ans[key] = best_path
+    return ans