Fix several things

bin/main.py

@@ -30,11 +30,13 @@
     LibriSpeechVocabulary,
 )
 from kospeech.dataclasses import (
-    FBankConfig,
+    FilterBankConfig,
     MelSpectrogramConfig,
-    MFCCConfig,
+    MfccConfig,
     SpectrogramConfig,
-    TrainConfig,
+    DeepSpeech2TrainConfig,
+    ListenAttendSpellTrainConfig,
+    TransformerTrainConfig,
     DeepSpeech2Config,
     JointCTCAttentionLASConfig,
     ListenAttendSpellConfig,
@@ -51,9 +53,11 @@ def train(config: DictConfig):
 
     if config.train.dataset == 'kspon':
         if config.train.output_unit == 'subword':
-            vocab = KsponSpeechVocabulary(vocab_path='../../../data/vocab/kspon_sentencepiece.vocab',
-                                          output_unit=config.train.output_unit,
-                                          sp_model_path='../../../data/vocab/kspon_sentencepiece.model')
+            vocab = KsponSpeechVocabulary(
+                vocab_path='../../../data/vocab/kspon_sentencepiece.vocab',
+                output_unit=config.train.output_unit,
+                sp_model_path='../../../data/vocab/kspon_sentencepiece.model',
+            )
         else:
             vocab = KsponSpeechVocabulary(
                 f'../../../data/vocab/aihub_{config.train.output_unit}_vocabs.csv', output_unit=config.train.output_unit
@@ -122,13 +126,13 @@ def train(config: DictConfig):
 
 
 cs = ConfigStore.instance()
-cs.store(group="audio", name="fbank", node=FBankConfig, package="audio")
+cs.store(group="audio", name="fbank", node=FilterBankConfig, package="audio")
 cs.store(group="audio", name="melspectrogram", node=MelSpectrogramConfig, package="audio")
-cs.store(group="audio", name="mfcc", node=MFCCConfig, package="audio")
+cs.store(group="audio", name="mfcc", node=MfccConfig, package="audio")
 cs.store(group="audio", name="spectrogram", node=SpectrogramConfig, package="audio")
-cs.store(group="train", name="ds2_train", node=TrainConfig, package="train")
-cs.store(group="train", name="las_train", node=TrainConfig, package="train")
-cs.store(group="train", name="transformer_train", node=TrainConfig, package="train")
+cs.store(group="train", name="ds2_train", node=DeepSpeech2TrainConfig, package="train")
+cs.store(group="train", name="las_train", node=ListenAttendSpellTrainConfig, package="train")
+cs.store(group="train", name="transformer_train", node=TransformerTrainConfig, package="train")
 cs.store(group="model", name="ds2", node=DeepSpeech2Config, package="model")
 cs.store(group="model", name="las", node=ListenAttendSpellConfig, package="model")
 cs.store(group="model", name="transformer", node=TransformerConfig, package="model")

configs/train.yaml

@@ -1,4 +1,4 @@
 defaults:
   - audio: fbank
-  - model: joint-ctc-attention-las
-  - train: las_train
+  - model: joint-ctc-attention-transformer
+  - train: transformer_train

kospeech/checkpoint/checkpoint.py

@@ -29,13 +29,11 @@ class Checkpoint(object):
         epoch (int): current epoch (an epoch is a loop through the full training data)
 
     Attributes:
-        CHECKPOINT_DIR_NAME (str): name of the checkpoint directory
         SAVE_PATH (str): path of file to save
         TRAINER_STATE_NAME (str): name of the file storing trainer states
         MODEL_NAME (str): name of the file storing model
     """
 
-    CHECKPOINT_DIR_NAME = 'checkpoints'
     SAVE_PATH = 'outputs'
     TRAINER_STATE_NAME = 'trainer_states.pt'
     MODEL_NAME = 'model.pt'

kospeech/dataclasses/__init__.py

@@ -0,0 +1,25 @@
+# -*- coding: utf-8 -*-
+# Soohwan Kim, Seyoung Bae, Cheolhwang Won.
+# @ArXiv : KoSpeech: Open-Source Toolkit for End-to-End Korean Speech Recognition
+# This source code is licensed under the Apache 2.0 License license found in the
+# LICENSE file in the root directory of this source tree.
+
+from .evaluate import EvalConfig
+from .audio import (
+    MfccConfig,
+    MelSpectrogramConfig,
+    SpectrogramConfig,
+    FilterBankConfig,
+)
+from .model import (
+    DeepSpeech2Config,
+    ListenAttendSpellConfig,
+    TransformerConfig,
+    JointCTCAttentionLASConfig,
+    JointCTCAttentionTransformerConfig,
+)
+from .train import (
+    DeepSpeech2TrainConfig,
+    ListenAttendSpellTrainConfig,
+    TransformerTrainConfig,
+)

kospeech/dataclasses/audio.py

@@ -0,0 +1,50 @@
+# -*- coding: utf-8 -*-
+# Soohwan Kim, Seyoung Bae, Cheolhwang Won.
+# @ArXiv : KoSpeech: Open-Source Toolkit for End-to-End Korean Speech Recognition
+# This source code is licensed under the Apache 2.0 License license found in the
+# LICENSE file in the root directory of this source tree.
+
+from dataclasses import dataclass
+
+
+@dataclass
+class AudioConfig:
+    audio_extension: str = "pcm"
+    sample_rate: int = 16000
+    frame_length: int = 20
+    frame_shift: int = 10
+    normalize: bool = True
+    del_silence: bool = True
+    feature_extract_by: str = "kaldi"
+    time_mask_num: int = 4
+    freq_mask_num: int = 2
+    spec_augment: bool = True
+    input_reverse: bool = False
+
+
+@dataclass
+class FilterBankConfig(AudioConfig):
+    transform_method: str = "fbank"
+    n_mels: int = 80
+    freq_mask_para: int = 18
+
+
+@dataclass
+class MelSpectrogramConfig(AudioConfig):
+    transform_method: str = "mel"
+    n_mels: int = 80
+    freq_mask_para: int = 18
+
+
+@dataclass
+class MfccConfig(AudioConfig):
+    transform_method: str = "mfcc"
+    n_mels: int = 40
+    freq_mask_para: int = 8
+
+
+@dataclass
+class SpectrogramConfig(AudioConfig):
+    transform_method: str = "spectrogram"
+    n_mels: int = 161  # Not used
+    freq_mask_para: int = 24

kospeech/dataclasses/evaluate.py

@@ -0,0 +1,22 @@
+# -*- coding: utf-8 -*-
+# Soohwan Kim, Seyoung Bae, Cheolhwang Won.
+# @ArXiv : KoSpeech: Open-Source Toolkit for End-to-End Korean Speech Recognition
+# This source code is licensed under the Apache 2.0 License license found in the
+# LICENSE file in the root directory of this source tree.
+
+from dataclasses import dataclass
+
+
+@dataclass
+class EvalConfig:
+    dataset: str = 'kspon'
+    dataset_path: str = ''
+    transcript_path: str = '../../../data/eval_transcript.txt'
+    model_path: str = ''
+    output_unit: str = 'character'
+    batch_size: int = 32
+    num_workers: int = 4
+    print_every: int = 20
+    decode: str = 'greedy'
+    k: int = 3
+    use_cuda: bool = True