tokenization within datasets (need to re-run json prep stage right

    before training stage to be compatible with this change) (#58)

espresso/criterions/cross_entropy_v2.py

@@ -72,9 +72,8 @@ def forward(self, model, sample, reduce=True):
             assert pred.size() == target.size()
             with data_utils.numpy_seed(model.num_updates):
                 i = np.random.randint(0, len(sample["id"]))
-            ref_tokens = sample["target_raw_text"][i]
             length = utils.strip_pad(target.data[i], self.padding_idx).size(0)
-            ref_one = self.dictionary.wordpiece_decode(ref_tokens)
+            ref_one = sample["text"][i]
             pred_one = self.dictionary.wordpiece_decode(self.dictionary.string(pred.data[i][:length]))
             logger.info("sample REF: " + ref_one)
             logger.info("sample PRD: " + pred_one)

espresso/criterions/label_smoothed_cross_entropy_v2.py

@@ -177,9 +177,8 @@ def forward(self, model, sample, reduce=True):
             assert pred.size() == target.size()
             with data_utils.numpy_seed(model.num_updates):
                 i = np.random.randint(0, len(sample["id"]))
-            ref_tokens = sample["target_raw_text"][i]
             length = utils.strip_pad(target.data[i], self.padding_idx).size(0)
-            ref_one = self.dictionary.wordpiece_decode(ref_tokens)
+            ref_one = sample["text"][i]
             pred_one = self.dictionary.wordpiece_decode(self.dictionary.string(pred.data[i][:length]))
             logger.info("sample REF: " + ref_one)
             logger.info("sample PRD: " + pred_one)

espresso/data/asr_chain_dataset.py

@@ -283,7 +283,7 @@ def get_batch_shapes(self):
 
     def __getitem__(self, index):
         tgt_item = self.tgt[index] if self.tgt is not None else None
-        text_item = self.text[index][1] if self.text is not None else None
+        text_item = self.text[index][2] if self.text is not None else None
         src_item = self.src[index]
         example = {
             "id": index,

espresso/data/asr_dataset.py

@@ -92,9 +92,13 @@ def merge(key, left_pad, move_eos_to_beginning=False, pad_to_length=None):
     else:
         ntokens = src_lengths.sum().item()
 
-    target_raw_text = None
-    if samples[0].get("target_raw_text", None) is not None:
-        target_raw_text = [samples[i]["target_raw_text"] for i in sort_order.numpy()]
+    token_text = None
+    if samples[0].get("token_text", None) is not None:
+        token_text = [samples[i]["token_text"] for i in sort_order.numpy()]
+
+    text = None
+    if samples[0].get("text", None) is not None:
+        text = [samples[i]["text"] for i in sort_order.numpy()]
 
     batch = {
         "id": id,
@@ -103,7 +107,8 @@ def merge(key, left_pad, move_eos_to_beginning=False, pad_to_length=None):
         "ntokens": ntokens,
         "net_input": {"src_tokens": src_frames, "src_lengths": src_lengths},
         "target": target,
-        "target_raw_text": target_raw_text,
+        "token_text": token_text,
+        "text": text,
     }
     if prev_output_tokens is not None:
         batch["net_input"]["prev_output_tokens"] = prev_output_tokens.index_select(
@@ -253,14 +258,16 @@ def get_batch_shapes(self):
 
     def __getitem__(self, index):
         tgt_item = self.tgt[index][0] if self.tgt is not None else None
-        raw_text_item = self.tgt[index][1] if self.tgt is not None else None
+        token_text_item = self.tgt[index][1] if self.tgt is not None else None
+        text_item = self.tgt[index][2] if self.tgt is not None else None
         src_item = self.src[index]
         example = {
             "id": index,
             "utt_id": self.src.utt_ids[index],
             "source": src_item,
             "target": tgt_item,
-            "target_raw_text": raw_text_item,
+            "token_text": token_text_item,
+            "text": text_item,
         }
         if self.constraints is not None:
             example["constraints"] = self.constraints[index]
@@ -304,7 +311,8 @@ def collater(self, samples, pad_to_length=None):
                 - `target` (LongTensor): a padded 2D Tensor of tokens in the
                   target sentence of shape `(bsz, tgt_len)`. Padding will appear
                   on the left if *left_pad_target* is ``True``.
-                - `target_raw_text` (List[str]): list of original text
+                - `token_text` (List[str]): list of token text
+                - `text` (List[str]): list of original text
                 - `tgt_lang_id` (LongTensor): a long Tensor which contains target language
                   IDs of each sample in the batch
         """

espresso/data/asr_dictionary.py

@@ -8,8 +8,8 @@
 
 import torch
 from fairseq.data import Dictionary, encoders
-from fairseq.dataclass import FairseqDataclass
 from fairseq.file_io import PathManager
+from omegaconf import DictConfig
 
 # will automatically load modules defined from there
 from espresso.data import encoders as encoders_espresso
@@ -99,12 +99,12 @@ def dummy_sentence(self, length):
         t[-1] = self.eos()
         return t
 
-    def build_tokenizer(self, cfg: Union[FairseqDataclass, Namespace]):
+    def build_tokenizer(self, cfg: Union[DictConfig, Namespace]):
         self.tokenizer = encoders.build_tokenizer(cfg)
 
-    def build_bpe(self, cfg: Union[FairseqDataclass, Namespace]):
+    def build_bpe(self, cfg: Union[DictConfig, Namespace]):
         if (
-            (isinstance(cfg, FairseqDataclass) and cfg._name == "characters_asr")
+            (isinstance(cfg, DictConfig) and cfg._name == "characters_asr")
             or (isinstance(cfg, Namespace) and getattr(cfg, "bpe", None) == "characters_asr")
         ):
             self.bpe = encoders.build_bpe(

espresso/data/asr_xent_dataset.py

@@ -524,7 +524,7 @@ def get_batch_shapes(self):
 
     def __getitem__(self, index):
         tgt_item = self.tgt[index] if self.tgt is not None else None
-        text_item = self.text[index][1] if self.text is not None else None
+        text_item = self.text[index][2] if self.text is not None else None
         src_item = self.src[index]
         example = {
             "id": index,

espresso/data/feat_text_dataset.py

@@ -233,38 +233,39 @@ def __getitem__(self, i):
 
 
 class AsrTextDataset(torch.utils.data.Dataset):
-    """Takes a text file as input and binarizes it in memory at instantiation.
-    Original lines are also kept in memory. Each line of the text file is in the
-    format of 'utt_id tokenized_text'."""
+    """Takes a text file as input, tokenizes and tensorizes it in memory at instantiation.
+    Both original text and tokenized text are kept in memory."""
 
-    def __init__(self, utt_ids: List[str], token_text: List[str], dictionary=None, append_eos=True):
+    def __init__(self, utt_ids: List[str], texts: List[str], dictionary=None, append_eos=True):
         super().__init__()
         self.dtype = np.float
         self.append_eos = append_eos
-        self.read_text(utt_ids, token_text, dictionary)
+        self.read_text(utt_ids, texts, dictionary)
 
-    def read_text(self, utt_ids: List[str], token_text: List[str], dictionary=None):
-        assert len(utt_ids) == len(token_text)
+    def read_text(self, utt_ids: List[str], texts: List[str], dictionary=None):
+        assert len(utt_ids) == len(texts)
         self.utt_ids = utt_ids
-        self.tokens_list = token_text
-        self.tensor_list = []
+        self.texts = texts
         self.size = len(self.utt_ids)  # number of utterances
-        self.sizes = []
+        self.token_texts = None
+        self.tensor_list = None
         if dictionary is not None:
-            for tokens in self.tokens_list:
-                tensor = dictionary.encode_line(
-                    tokens, add_if_not_exist=False, append_eos=self.append_eos,
-                ).long()
-                self.tensor_list.append(tensor)
-                self.sizes.append(len(self.tensor_list[-1]))
+            self.token_texts = [dictionary.wordpiece_encode(x) for x in texts]
+            self.tensor_list = [
+                dictionary.encode_line(tokens, add_if_not_exist=False, append_eos=self.append_eos).long()
+                for tokens in self.token_texts
+            ]
+            self.sizes = [len(tensor) for tensor in self.tensor_list]
         else:
-            self.sizes = [len(tokenize_line(tokens)) for tokens in self.tokens_list]
+            self.sizes = [len(tokenize_line(text)) for text in texts]
 
         self.sizes = np.array(self.sizes, dtype=np.int32)
 
         assert (
-            len(self.utt_ids) == len(self.tokens_list)
-            and (dictionary is None or len(self.utt_ids) == len(self.tensor_list))
+            (
+                dictionary is None
+                or (len(self.utt_ids) == len(self.tensor_list) and len(self.utt_ids) == len(self.token_texts))
+            )
             and len(self.utt_ids) == len(self.sizes)
         )
 
@@ -280,15 +281,21 @@ def filter_and_reorder(self, indices):
             len(np.unique(indices)) == len(indices)
         ), "Duplicate elements in indices."
         self.utt_ids = [self.utt_ids[i] for i in indices]
-        self.tokens_list = [self.tokens_list[i] for i in indices]
-        if len(self.tensor_list) > 0:
+        self.texts = [self.texts[i] for i in indices]
+        if self.token_texts is not None:
+            self.token_texts = [self.token_texts[i] for i in indices]
+        if self.tensor_list is not None:
             self.tensor_list = [self.tensor_list[i] for i in indices]
         self.sizes = self.sizes[indices]
         self.size = len(self.utt_ids)
 
     def __getitem__(self, i):
         self.check_index(i)
-        return self.tensor_list[i] if len(self.tensor_list) > 0 else None, self.tokens_list[i]
+        return (
+            self.tensor_list[i] if self.tensor_list is not None else None,
+            self.token_texts[i] if self.token_texts is not None else None,
+            self.texts[i]
+        )
 
     def __len__(self):
         return self.size

espresso/speech_recognize.py

@@ -83,6 +83,8 @@ def _main(cfg, output_file):
     use_cuda = torch.cuda.is_available() and not cfg.common.cpu
 
     task = tasks.setup_task(cfg.task)
+    task.build_tokenizer(cfg.tokenizer)
+    task.build_bpe(cfg.bpe)
 
     # Set dictionary
     dictionary = task.target_dictionary
@@ -253,7 +255,7 @@ def decode_fn(x):
 
             # Retrieve the original sentences
             if has_target:
-                target_str = sample["target_raw_text"][i]
+                target_str = sample["token_text"][i]
                 if not cfg.common_eval.quiet:
                     detok_target_str = decode_fn(target_str)
                     print("T-{}\t{}".format(utt_id, detok_target_str), file=output_file)

espresso/tasks/speech_recognition.py

@@ -109,7 +109,7 @@ def get_asr_dataset_from_json(
     {
         "011c0202": {
             "feat": "fbank/raw_fbank_pitch_train_si284.1.ark:54819",
-            "token_text": "T H E <space> H O T E L",
+            "text": "THE HOTEL",
             "utt2num_frames": "693",
         },
         "011c0203": {
@@ -133,12 +133,12 @@ def get_asr_dataset_from_json(
         with open(data_json_path, "rb") as f:
             loaded_json = json.load(f, object_pairs_hook=OrderedDict)
 
-        utt_ids, feats, token_text, utt2num_frames = [], [], [], []
+        utt_ids, feats, texts, utt2num_frames = [], [], [], []
         for utt_id, val in loaded_json.items():
             utt_ids.append(utt_id)
             feats.append(val["feat"])
-            if "token_text" in val:
-                token_text.append(val["token_text"])
+            if "text" in val:
+                texts.append(val["text"])
             if "utt2num_frames" in val:
                 utt2num_frames.append(int(val["utt2num_frames"]))
 
@@ -148,10 +148,10 @@ def get_asr_dataset_from_json(
             specaugment_config=specaugment_config if split == "train" else None,
             ordered_prefetch=True,
         ))
-        if len(token_text) > 0:
-            assert len(utt_ids) == len(token_text)
+        if len(texts) > 0:
+            assert len(utt_ids) == len(texts)
             assert tgt_dict is not None
-            tgt_datasets.append(AsrTextDataset(utt_ids, token_text, tgt_dict))
+            tgt_datasets.append(AsrTextDataset(utt_ids, texts, tgt_dict))
 
         logger.info("{} {} examples".format(data_json_path, len(src_datasets[-1])))
 
@@ -196,7 +196,7 @@ def get_asr_dataset_from_json(
 @register_task("speech_recognition_espresso", dataclass=SpeechRecognitionEspressoConfig)
 class SpeechRecognitionEspressoTask(FairseqTask):
     """
-    Transcribe from speech (source) to token text (target).
+    Transcribe from speech (source) to text (target).
 
     Args:
         tgt_dict (~fairseq.data.AsrDictionary): dictionary for the output tokens
@@ -406,7 +406,7 @@ def _inference_with_wer(self, decoder, sample, model):
         scorer.reset()
         for i in range(target.size(0)):
             utt_id = sample["utt_id"][i]
-            ref_tokens = sample["target_raw_text"][i]
+            ref_tokens = sample["token_text"][i]
             pred_tokens = self.target_dictionary.string(pred.data[i])
             scorer.add_evaluation(utt_id, ref_tokens, pred_tokens)
         return (

examples/asr_librispeech/run.sh

@@ -204,18 +204,18 @@ fi
 if [ ${stage} -le 7 ]; then
   echo "Stage 7: Dump Json Files"
   train_feat=$train_feat_dir/feats.scp
-  train_token_text=data/$train_set/token_text
+  train_text=data/$train_set/text
   train_utt2num_frames=data/$train_set/utt2num_frames
   valid_feat=$valid_feat_dir/feats.scp
-  valid_token_text=data/$valid_set/token_text
+  valid_text=data/$valid_set/text
   valid_utt2num_frames=data/$valid_set/utt2num_frames
-  asr_prep_json.py --feat-files $train_feat --token-text-files $train_token_text --utt2num-frames-files $train_utt2num_frames --output data/train.json
-  asr_prep_json.py --feat-files $valid_feat --token-text-files $valid_token_text --utt2num-frames-files $valid_utt2num_frames --output data/valid.json
+  asr_prep_json.py --feat-files $train_feat --text-files $train_text --utt2num-frames-files $train_utt2num_frames --output data/train.json
+  asr_prep_json.py --feat-files $valid_feat --text-files $valid_text --utt2num-frames-files $valid_utt2num_frames --output data/valid.json
   for dataset in $test_set; do
     feat=${dumpdir}/$dataset/delta${do_delta}/feats.scp
-    token_text=data/$dataset/token_text
+    text=data/$dataset/text
     utt2num_frames=data/$dataset/utt2num_frames
-    asr_prep_json.py --feat-files $feat --token-text-files $token_text --utt2num-frames-files $utt2num_frames --output data/$dataset.json
+    asr_prep_json.py --feat-files $feat --text-files $text --utt2num-frames-files $utt2num_frames --output data/$dataset.json
   done
 fi
 

examples/asr_swbd/run.sh

@@ -242,18 +242,18 @@ fi
 if [ $stage -le 6 ]; then
   echo "Stage 6: Dump Json Files"
   train_feat=$train_feat_dir/feats.scp
-  train_token_text=data/$train_set/token_text
+  train_text=data/$train_set/text
   train_utt2num_frames=data/$train_set/utt2num_frames
   valid_feat=$valid_feat_dir/feats.scp
-  valid_token_text=data/$valid_set/token_text
+  valid_text=data/$valid_set/text
   valid_utt2num_frames=data/$valid_set/utt2num_frames
-  asr_prep_json.py --feat-files $train_feat --token-text-files $train_token_text --utt2num-frames-files $train_utt2num_frames --output data/train.json
-  asr_prep_json.py --feat-files $valid_feat --token-text-files $valid_token_text --utt2num-frames-files $valid_utt2num_frames --output data/valid.json
+  asr_prep_json.py --feat-files $train_feat --text-files $train_text --utt2num-frames-files $train_utt2num_frames --output data/train.json
+  asr_prep_json.py --feat-files $valid_feat --text-files $valid_text --utt2num-frames-files $valid_utt2num_frames --output data/valid.json
   for dataset in $test_set; do
     feat=${dumpdir}/$dataset/delta${do_delta}/feats.scp
     utt2num_frames=data/$dataset/utt2num_frames
     # only score train_dev with built-in scorer
-    text_opt= && [ "$dataset" == "train_dev" ] && text_opt="--token-text-files data/$dataset/token_text"
+    text_opt= && [ "$dataset" == "train_dev" ] && text_opt="--text-files data/$dataset/text"
     asr_prep_json.py --feat-files $feat $text_opt --utt2num-frames-files $utt2num_frames --output data/$dataset.json
   done
 fi

examples/asr_wsj/run.sh

@@ -247,22 +247,22 @@ fi
 if [ ${stage} -le 8 ]; then
   echo "Stage 8: Dump Json Files"
   train_feat=$train_feat_dir/feats.scp
-  train_token_text=data/$train_set/token_text
+  train_text=data/$train_set/text
   train_utt2num_frames=data/$train_set/utt2num_frames
   valid_feat=$valid_feat_dir/feats.scp
-  valid_token_text=data/$valid_set/token_text
+  valid_text=data/$valid_set/text
   valid_utt2num_frames=data/$valid_set/utt2num_frames
-  asr_prep_json.py --feat-files $train_feat --token-text-files $train_token_text --utt2num-frames-files $train_utt2num_frames --output data/train.json
-  asr_prep_json.py --feat-files $valid_feat --token-text-files $valid_token_text --utt2num-frames-files $valid_utt2num_frames --output data/valid.json
+  asr_prep_json.py --feat-files $train_feat --text-files $train_text --utt2num-frames-files $train_utt2num_frames --output data/train.json
+  asr_prep_json.py --feat-files $valid_feat --text-files $valid_text --utt2num-frames-files $valid_utt2num_frames --output data/valid.json
   for dataset in $valid_set $test_set; do
     if [ "$dataset" == "$valid_set" ]; then
       feat=$valid_feat_dir/feats.scp
     elif [ "$dataset" == "$test_set" ]; then
       feat=$test_feat_dir/feats.scp
     fi
-    token_text=data/$dataset/token_text
+    text=data/$dataset/text
     utt2num_frames=data/$dataset/utt2num_frames
-    asr_prep_json.py --feat-files $feat --token-text-files $token_text --utt2num-frames-files $utt2num_frames --output data/$dataset.json
+    asr_prep_json.py --feat-files $feat --text-files $text --utt2num-frames-files $utt2num_frames --output data/$dataset.json
   done
 fi