Update beam search to support max/log_add in selecting duplicate hyps.

egs/librispeech/ASR/transducer_stateless/beam_search.py

@@ -21,6 +21,153 @@ import k2
 import torch
 from model import Transducer
 
+from icefall.decode import Nbest, one_best_decoding
+from icefall.utils import get_texts
+
+
+def fast_beam_search(
+    model: Transducer,
+    decoding_graph: k2.Fsa,
+    encoder_out: torch.Tensor,
+    encoder_out_lens: torch.Tensor,
+    beam: float,
+    max_states: int,
+    max_contexts: int,
+    use_max: bool = False,
+) -> List[List[int]]:
+    """It limits the maximum number of symbols per frame to 1.
+
+    Args:
+      model:
+        An instance of `Transducer`.
+      decoding_graph:
+        Decoding graph used for decoding, may be a TrivialGraph or a HLG.
+      encoder_out:
+        A tensor of shape (N, T, C) from the encoder.
+      encoder_out_lens:
+        A tensor of shape (N,) containing the number of frames in `encoder_out`
+        before padding.
+      beam:
+        Beam value, similar to the beam used in Kaldi..
+      max_states:
+        Max states per stream per frame.
+      max_contexts:
+        Max contexts pre stream per frame.
+      use_max:
+        True to use max operation to select the hypothesis with the largest
+        log_prob when there are duplicate hypotheses; False to use log-add.
+    Returns:
+      Return the decoded result.
+    """
+    assert encoder_out.ndim == 3
+
+    context_size = model.decoder.context_size
+    vocab_size = model.decoder.vocab_size
+
+    B, T, C = encoder_out.shape
+
+    config = k2.RnntDecodingConfig(
+        vocab_size=vocab_size,
+        decoder_history_len=context_size,
+        beam=beam,
+        max_contexts=max_contexts,
+        max_states=max_states,
+    )
+    individual_streams = []
+    for i in range(B):
+        individual_streams.append(k2.RnntDecodingStream(decoding_graph))
+    decoding_streams = k2.RnntDecodingStreams(individual_streams, config)
+
+    tmp_len = torch.tensor([1])
+
+    for t in range(T):
+        # shape is a RaggedShape of shape (B, context)
+        # contexts is a Tensor of shape (shape.NumElements(), context_size)
+        shape, contexts = decoding_streams.get_contexts()
+        # `nn.Embedding()` in torch below v1.7.1 supports only torch.int64
+        contexts = contexts.to(torch.int64)
+        # decoder_out is of shape (shape.NumElements(), 1, decoder_out_dim)
+        decoder_out = model.decoder(contexts, need_pad=False)
+        # current_encoder_out is of shape
+        # (shape.NumElements(), 1, encoder_out_dim)
+        # fmt: off
+        current_encoder_out = torch.index_select(
+            encoder_out[:, t:t + 1, :], 0, shape.row_ids(1)
+        )
+        # fmt: on
+        logits = model.joiner(
+            current_encoder_out,
+            decoder_out,
+            tmp_len.expand(decoder_out.size(0)),
+            tmp_len.expand(decoder_out.size(0)),
+        )
+        log_probs = logits.log_softmax(dim=-1)
+        decoding_streams.advance(log_probs)
+    decoding_streams.terminate_and_flush_to_streams()
+    lattice = decoding_streams.format_output(encoder_out_lens.tolist())
+
+    if use_max:
+        best_path = one_best_decoding(lattice)
+        hyps = get_texts(best_path)
+        return hyps
+    else:
+        num_paths = 20
+        use_double_scores = True
+        nbest_scale = 0.8
+
+        nbest = Nbest.from_lattice(
+            lattice=lattice,
+            num_paths=num_paths,
+            use_double_scores=use_double_scores,
+            nbest_scale=nbest_scale,
+        )
+        # The following code is modified from nbest.intersect()
+        word_fsa = k2.invert(nbest.fsa)
+        if hasattr(lattice, "aux_labels"):
+            # delete token IDs as it is not needed
+            del word_fsa.aux_labels
+        word_fsa.scores.zero_()
+        word_fsa_with_epsilon_loops = k2.remove_epsilon_and_add_self_loops(
+            word_fsa
+        )
+        path_to_utt_map = nbest.shape.row_ids(1)
+
+        if hasattr(lattice, "aux_labels"):
+            # lattice has token IDs as labels and word IDs as aux_labels.
+            # inv_lattice has word IDs as labels and token IDs as aux_labels
+            inv_lattice = k2.invert(lattice)
+            inv_lattice = k2.arc_sort(inv_lattice)
+        else:
+            inv_lattice = k2.arc_sort(lattice)
+
+        if inv_lattice.shape[0] == 1:
+            path_lattice = k2.intersect_device(
+                inv_lattice,
+                word_fsa_with_epsilon_loops,
+                b_to_a_map=torch.zeros_like(path_to_utt_map),
+                sorted_match_a=True,
+            )
+        else:
+            path_lattice = k2.intersect_device(
+                inv_lattice,
+                word_fsa_with_epsilon_loops,
+                b_to_a_map=path_to_utt_map,
+                sorted_match_a=True,
+            )
+
+        # path_lattice has word IDs as labels and token IDs as aux_labels
+        path_lattice = k2.top_sort(k2.connect(path_lattice))
+
+        tot_scores = path_lattice.get_tot_scores(
+            use_double_scores=True, log_semiring=True
+        )
+        ragged_tot_scores = k2.RaggedTensor(nbest.shape, tot_scores)
+        best_hyp_indexes = ragged_tot_scores.argmax()
+
+        best_path = k2.index_fsa(nbest.fsa, best_hyp_indexes)
+        hyps = get_texts(best_path)
+        return hyps
+
 
 def greedy_search(
     model: Transducer, encoder_out: torch.Tensor, max_sym_per_frame: int
@@ -203,7 +350,7 @@ class HypothesisList(object):
     def data(self) -> Dict[str, Hypothesis]:
         return self._data
 
-    def add(self, hyp: Hypothesis) -> None:
+    def add(self, hyp: Hypothesis, use_max: bool = False) -> None:
         """Add a Hypothesis to `self`.
 
         If `hyp` already exists in `self`, its probability is updated using
@@ -212,13 +359,20 @@ class HypothesisList(object):
         Args:
           hyp:
             The hypothesis to be added.
+          use_max:
+            True to select the hypothesis with the larger log_prob in case there
+            already exists a hypothesis whose `ys` equals to `hyp.ys`.
+            False to use log_add.
         """
         key = hyp.key
         if key in self:
             old_hyp = self._data[key]  # shallow copy
-            torch.logaddexp(
+            if use_max:
-                old_hyp.log_prob, hyp.log_prob, out=old_hyp.log_prob
+                old_hyp.log_prob = max(old_hyp.log_prob, hyp.log_prob)
-            )
+            else:
+                torch.logaddexp(
+                    old_hyp.log_prob, hyp.log_prob, out=old_hyp.log_prob
+                )
         else:
             self._data[key] = hyp
 
@@ -415,6 +569,7 @@ def modified_beam_search(
     model: Transducer,
     encoder_out: torch.Tensor,
     beam: int = 4,
+    use_max: bool = False,
 ) -> List[List[int]]:
     """Beam search in batch mode with --max-sym-per-frame=1 being hardcodded.
 
@@ -425,6 +580,10 @@ def modified_beam_search(
         Output from the encoder. Its shape is (N, T, C).
       beam:
         Number of active paths during the beam search.
+      use_max:
+        If True, it uses max operation to select the hypothesis with the
+        larger log_prob in case two hypotheses have the same token sequences.
+        If False, use log add.
     Returns:
       Return a list-of-list of token IDs. ans[i] is the decoding results
       for the i-th utterance.
@@ -443,7 +602,8 @@ def modified_beam_search(
             Hypothesis(
                 ys=[blank_id] * context_size,
                 log_prob=torch.zeros(1, dtype=torch.float32, device=device),
-            )
+            ),
+            use_max=use_max,
         )
 
     encoder_out_len = torch.tensor([1])
@@ -519,7 +679,7 @@ def modified_beam_search(
 
                 new_log_prob = topk_log_probs[k]
                 new_hyp = Hypothesis(ys=new_ys, log_prob=new_log_prob)
-                B[i].add(new_hyp)
+                B[i].add(new_hyp, use_max=use_max)
 
     best_hyps = [b.get_most_probable(length_norm=True) for b in B]
     ans = [h.ys[context_size:] for h in best_hyps]
@@ -531,6 +691,7 @@ def _deprecated_modified_beam_search(
     model: Transducer,
     encoder_out: torch.Tensor,
     beam: int = 4,
+    use_max: bool = False,
 ) -> List[int]:
     """It limits the maximum number of symbols per frame to 1.
 
@@ -545,6 +706,10 @@ def _deprecated_modified_beam_search(
         A tensor of shape (N, T, C) from the encoder. Support only N==1 for now.
       beam:
         Beam size.
+      use_max:
+        If True, it uses max operation to select the hypothesis with the
+        larger log_prob in case two hypotheses have the same token sequences.
+        If False, use log add.
     Returns:
       Return the decoded result.
     """
@@ -565,7 +730,8 @@ def _deprecated_modified_beam_search(
         Hypothesis(
             ys=[blank_id] * context_size,
             log_prob=torch.zeros(1, dtype=torch.float32, device=device),
-        )
+        ),
+        use_max=use_max,
     )
 
     encoder_out_len = torch.tensor([1])
@@ -624,7 +790,7 @@ def _deprecated_modified_beam_search(
                 new_ys.append(new_token)
             new_log_prob = topk_log_probs[i]
             new_hyp = Hypothesis(ys=new_ys, log_prob=new_log_prob)
-            B.add(new_hyp)
+            B.add(new_hyp, use_max=use_max)
 
     best_hyp = B.get_most_probable(length_norm=True)
     ys = best_hyp.ys[context_size:]  # [context_size:] to remove blanks
@@ -636,6 +802,7 @@ def beam_search(
     model: Transducer,
     encoder_out: torch.Tensor,
     beam: int = 4,
+    use_max: bool = False,
 ) -> List[int]:
     """
     It implements Algorithm 1 in https://arxiv.org/pdf/1211.3711.pdf
@@ -649,6 +816,10 @@ def beam_search(
         A tensor of shape (N, T, C) from the encoder. Support only N==1 for now.
       beam:
         Beam size.
+      use_max:
+        If True, it uses max operation to select the hypothesis with the
+        larger log_prob in case two hypotheses have the same token sequences.
+        If False, use log add.
     Returns:
       Return the decoded result.
     """
@@ -675,7 +846,8 @@ def beam_search(
         Hypothesis(
             ys=[blank_id] * context_size,
             log_prob=torch.zeros(1, dtype=torch.float32, device=device),
-        )
+        ),
+        use_max=use_max,
     )
 
     max_sym_per_utt = 20000
@@ -721,7 +893,10 @@ def beam_search(
             new_y_star_log_prob = y_star.log_prob + skip_log_prob
 
             # ys[:] returns a copy of ys
-            B.add(Hypothesis(ys=y_star.ys[:], log_prob=new_y_star_log_prob))
+            B.add(
+                Hypothesis(ys=y_star.ys[:], log_prob=new_y_star_log_prob),
+                use_max=use_max,
+            )
 
             # Second, process other non-blank labels
             values, indices = log_prob.topk(beam + 1)
@@ -733,7 +908,10 @@ def beam_search(
                 new_ys = y_star.ys + [i]
 
                 new_log_prob = y_star.log_prob + values[idx]
-                A.add(Hypothesis(ys=new_ys, log_prob=new_log_prob))
+                A.add(
+                    Hypothesis(ys=new_ys, log_prob=new_log_prob),
+                    use_max=use_max,
+                )
 
             # Check whether B contains more than "beam" elements more probable
             # than the most probable in A

egs/librispeech/ASR/transducer_stateless/decode.py

@@ -22,7 +22,8 @@ Usage:
         --epoch 14 \
         --avg 7 \
         --exp-dir ./transducer_stateless/exp \
-        --max-duration 100 \
+        --max-duration 1000 \
+        --max-sym-per-frame 1 \
         --decoding-method greedy_search
 
 (2) beam search
@@ -30,7 +31,7 @@ Usage:
         --epoch 14 \
         --avg 7 \
         --exp-dir ./transducer_stateless/exp \
-        --max-duration 100 \
+        --max-duration 1000 \
         --decoding-method beam_search \
         --beam-size 4
 
@@ -39,9 +40,20 @@ Usage:
         --epoch 14 \
         --avg 7 \
         --exp-dir ./transducer_stateless/exp \
-        --max-duration 100 \
+        --max-duration 1000 \
         --decoding-method modified_beam_search \
         --beam-size 4
+
+(4) fast beam search
+./transducer_stateless/decode.py \
+        --epoch 14 \
+        --avg 7 \
+        --exp-dir ./transducer_stateless/exp \
+        --max-duration 1000 \
+        --decoding-method fast_beam_search \
+        --beam 4 \
+        --max-contexts 4 \
+        --max-states 8
 """
 
 
@@ -49,14 +61,16 @@ import argparse
 import logging
 from collections import defaultdict
 from pathlib import Path
-from typing import Dict, List, Tuple
+from typing import Dict, List, Optional, Tuple
 
+import k2
 import sentencepiece as spm
 import torch
 import torch.nn as nn
 from asr_datamodule import LibriSpeechAsrDataModule
 from beam_search import (
     beam_search,
+    fast_beam_search,
     greedy_search,
     greedy_search_batch,
     modified_beam_search,
@@ -68,6 +82,7 @@ from icefall.utils import (
     AttributeDict,
     setup_logger,
     store_transcripts,
+    str2bool,
     write_error_stats,
 )
 
@@ -115,6 +130,7 @@ def get_parser():
           - greedy_search
           - beam_search
           - modified_beam_search
+          - fast_beam_search
         """,
     )
 
@@ -126,6 +142,32 @@ def get_parser():
         beam_search or modified_beam_search""",
     )
 
+    parser.add_argument(
+        "--beam",
+        type=float,
+        default=4,
+        help="""A floating point value to calculate the cutoff score during beam
+        search (i.e., `cutoff = max-score - beam`), which is the same as the
+        `beam` in Kaldi.
+        Used only when --decoding-method is fast_beam_search""",
+    )
+
+    parser.add_argument(
+        "--max-contexts",
+        type=int,
+        default=4,
+        help="""Used only when --decoding-method is
+        fast_beam_search""",
+    )
+
+    parser.add_argument(
+        "--max-states",
+        type=int,
+        default=8,
+        help="""Used only when --decoding-method is
+        fast_beam_search""",
+    )
+
     parser.add_argument(
         "--context-size",
         type=int,
@@ -141,6 +183,17 @@ def get_parser():
         Used only when --decoding_method is greedy_search""",
     )
 
+    parser.add_argument(
+        "--use-max",
+        type=str2bool,
+        default=False,
+        help="""If True, use max-op to select the hypothesis that have the
+        max log_prob in case of duplicate hypotheses.
+        If False, use log_add.
+        Used only for beam_search, modified_beam_search, and fast_beam_search
+        """,
+    )
+
     return parser
 
 
@@ -149,6 +202,7 @@ def decode_one_batch(
     model: nn.Module,
     sp: spm.SentencePieceProcessor,
     batch: dict,
+    decoding_graph: Optional[k2.Fsa] = None,
 ) -> Dict[str, List[List[str]]]:
     """Decode one batch and return the result in a dict. The dict has the
     following format:
@@ -190,7 +244,18 @@ def decode_one_batch(
     )
     hyp_list: List[List[int]] = []
 
-    if (
+    if params.decoding_method == "fast_beam_search":
+        hyp_list = fast_beam_search(
+            model=model,
+            decoding_graph=decoding_graph,
+            encoder_out=encoder_out,
+            encoder_out_lens=encoder_out_lens,
+            beam=params.beam,
+            max_contexts=params.max_contexts,
+            max_states=params.max_states,
+            use_max=params.use_max,
+        )
+    elif (
         params.decoding_method == "greedy_search"
         and params.max_sym_per_frame == 1
     ):
@@ -203,6 +268,7 @@ def decode_one_batch(
             model=model,
             encoder_out=encoder_out,
             beam=params.beam_size,
+            use_max=params.use_max,
         )
     else:
         batch_size = encoder_out.size(0)
@@ -221,6 +287,7 @@ def decode_one_batch(
                     model=model,
                     encoder_out=encoder_out_i,
                     beam=params.beam_size,
+                    use_max=params.use_max,
                 )
             else:
                 raise ValueError(
@@ -232,6 +299,14 @@ def decode_one_batch(
 
     if params.decoding_method == "greedy_search":
         return {"greedy_search": hyps}
+    elif params.decoding_method == "fast_beam_search":
+        return {
+            (
+                f"beam_{params.beam}_"
+                f"max_contexts_{params.max_contexts}_"
+                f"max_states_{params.max_states}"
+            ): hyps
+        }
     else:
         return {f"beam_{params.beam_size}": hyps}
 
@@ -241,6 +316,7 @@ def decode_dataset(
     params: AttributeDict,
     model: nn.Module,
     sp: spm.SentencePieceProcessor,
+    decoding_graph: Optional[k2.Fsa] = None,
 ) -> Dict[str, List[Tuple[List[str], List[str]]]]:
     """Decode dataset.
 
@@ -281,6 +357,7 @@ def decode_dataset(
             model=model,
             sp=sp,
             batch=batch,
+            decoding_graph=decoding_graph,
         )
 
         for name, hyps in hyps_dict.items():
@@ -358,15 +435,22 @@ def main():
     params.update(vars(args))
 
     assert params.decoding_method in (
-        "greedy_search",
         "beam_search",
+        "fast_beam_search",
+        "greedy_search",
         "modified_beam_search",
     )
     params.res_dir = params.exp_dir / params.decoding_method
 
     params.suffix = f"epoch-{params.epoch}-avg-{params.avg}"
-    if "beam_search" in params.decoding_method:
+    if "fast_beam_search" == params.decoding_method:
+        params.suffix += f"-beam-{params.beam}"
+        params.suffix += f"-max-contexts-{params.max_contexts}"
+        params.suffix += f"-max-states-{params.max_states}"
+        params.suffix += f"-use-max-{params.use_max}"
+    elif "beam_search" in params.decoding_method:
         params.suffix += f"-beam-{params.beam_size}"
+        params.suffix += f"-use-max-{params.use_max}"
     else:
         params.suffix += f"-context-{params.context_size}"
         params.suffix += f"-max-sym-per-frame-{params.max_sym_per_frame}"
@@ -408,6 +492,11 @@ def main():
     model.eval()
     model.device = device
 
+    if params.decoding_method == "fast_beam_search":
+        decoding_graph = k2.trivial_graph(params.vocab_size - 1, device=device)
+    else:
+        decoding_graph = None
+
     num_param = sum([p.numel() for p in model.parameters()])
     logging.info(f"Number of model parameters: {num_param}")
 
@@ -428,6 +517,7 @@ def main():
             params=params,
             model=model,
             sp=sp,
+            decoding_graph=decoding_graph,
         )
 
         save_results(

egs/librispeech/ASR/transducer_stateless/decoder.py

@@ -58,6 +58,7 @@ class Decoder(nn.Module):
             padding_idx=blank_id,
         )
         self.blank_id = blank_id
+        self.vocab_size = vocab_size
 
         assert context_size >= 1, context_size
         self.context_size = context_size

icefall/decode.py

@@ -218,7 +218,7 @@ class Nbest(object):
         # word_seq is a k2.RaggedTensor sharing the same shape as `path`
         # but it contains word IDs. Note that it also contains 0s and -1s.
         # The last entry in each sublist is -1.
-        # It axes is [utt][path][word_id]
+        # Its axes are [utt][path][word_id]
         if isinstance(lattice.aux_labels, torch.Tensor):
             word_seq = k2.ragged.index(lattice.aux_labels, path)
         else: