Merge branch 'develop' into patch-12

paddlespeech/audiotools/core/util.py

@@ -734,8 +734,8 @@ def default_collate(batch,
         if not all(len(elem) == elem_size for elem in it):
             raise RuntimeError(
                 "each element in list of batch should be of equal size")
-        transposed = list(zip(
+        transposed = list(
-            *batch))  # It may be accessed twice, so we use a list.
+            zip(*batch))  # It may be accessed twice, so we use a list.
 
         if isinstance(elem, tuple):
             return [

paddlespeech/audiotools/data/datasets.py

@@ -202,9 +202,9 @@ class AudioDataset:
 
     Examples
     --------
-    >>> from audio.audiotools.data.datasets import AudioLoader
+    >>> from paddlespeech.audiotools.data.datasets import AudioLoader
-    >>> from audio.audiotools.data.datasets import AudioDataset
+    >>> from paddlespeech.audiotools.data.datasets import AudioDataset
-    >>> from audio.audiotools import transforms as tfm
+    >>> from paddlespeech.audiotools import transforms as tfm
     >>> import numpy as np
     >>>
     >>> loaders = [
@@ -237,9 +237,9 @@ class AudioDataset:
 
     Below is an example of how one could load MUSDB multitrack data:
 
-    >>> from audio import audiotools as at
+    >>> from paddlespeech import audiotools as at
     >>> from pathlib import Path
-    >>> from audio.audiotools import transforms as tfm
+    >>> from paddlespeech.audiotools import transforms as tfm
     >>> import numpy as np
     >>> import torch
     >>>
@@ -296,9 +296,9 @@ class AudioDataset:
 
     Similarly, here's example code for loading Slakh data:
 
-    >>> from audio import audiotools as at
+    >>> from paddlespeech import audiotools as at
     >>> from pathlib import Path
-    >>> from audio.audiotools import transforms as tfm
+    >>> from paddlespeech.audiotools import transforms as tfm
     >>> import numpy as np
     >>> import torch
     >>> import glob

paddlespeech/audiotools/data/preprocess.py

@@ -37,7 +37,7 @@ def create_csv(audio_files: list,
 
     You can produce a CSV file from a directory of audio files via:
 
-    >>> from audio import audiotools
+    >>> from paddlespeech import audiotools
     >>> directory = ...
     >>> audio_files = audiotools.util.find_audio(directory)
     >>> output_path = "train.csv"

paddlespeech/audiotools/post.py

@@ -6,7 +6,7 @@ import typing
 
 import paddle
 
-from audio.audiotools.core import AudioSignal
+from paddlespeech.audiotools.core import AudioSignal
 
 
 def audio_table(

paddlespeech/audiotools/requirements.txt

@@ -2,5 +2,4 @@ ffmpeg-python
 ffmpy
 flatten_dict
 pyloudnorm
-pytest
+rich
-rich

paddlespeech/s2t/modules/conformer_convolution.py

@@ -19,7 +19,7 @@ from typing import Tuple
 import paddle
 from paddle import nn
 from paddle.nn import initializer as I
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 from paddlespeech.s2t.modules.align import BatchNorm1D
 from paddlespeech.s2t.modules.align import Conv1D
@@ -34,6 +34,7 @@ __all__ = ['ConvolutionModule']
 class ConvolutionModule(nn.Layer):
     """ConvolutionModule in Conformer model."""
 
+    @typechecked
     def __init__(self,
                  channels: int,
                  kernel_size: int=15,
@@ -52,7 +53,6 @@ class ConvolutionModule(nn.Layer):
             causal (bool): Whether use causal convolution or not
             bias (bool): Whether Conv with bias or not
         """
-        assert check_argument_types()
         super().__init__()
         self.bias = bias
         self.channels = channels

paddlespeech/s2t/modules/ctc.py

@@ -17,7 +17,7 @@ from typing import Union
 import paddle
 from paddle import nn
 from paddle.nn import functional as F
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 from paddlespeech.s2t.modules.align import Linear
 from paddlespeech.s2t.modules.loss import CTCLoss
@@ -48,6 +48,7 @@ __all__ = ['CTCDecoder']
 
 
 class CTCDecoderBase(nn.Layer):
+    @typechecked
     def __init__(self,
                  odim,
                  enc_n_units,
@@ -66,7 +67,6 @@ class CTCDecoderBase(nn.Layer):
             batch_average (bool): do batch dim wise average.
             grad_norm_type (str): Default, None. one of 'instance', 'batch', 'frame', None.
         """
-        assert check_argument_types()
         super().__init__()
 
         self.blank_id = blank_id

paddlespeech/s2t/modules/decoder.py

@@ -21,7 +21,7 @@ from typing import Tuple
 
 import paddle
 from paddle import nn
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 from paddlespeech.s2t.decoders.scorers.scorer_interface import BatchScorerInterface
 from paddlespeech.s2t.modules.align import Embedding
@@ -61,6 +61,7 @@ class TransformerDecoder(BatchScorerInterface, nn.Layer):
             False: x -> x + att(x)
     """
 
+    @typechecked
     def __init__(self,
                  vocab_size: int,
                  encoder_output_size: int,
@@ -77,8 +78,6 @@ class TransformerDecoder(BatchScorerInterface, nn.Layer):
                  concat_after: bool=False,
                  max_len: int=5000):
 
-        assert check_argument_types()
-
         nn.Layer.__init__(self)
         self.selfattention_layer_type = 'selfattn'
         attention_dim = encoder_output_size
@@ -276,6 +275,7 @@ class BiTransformerDecoder(BatchScorerInterface, nn.Layer):
             False: x -> x + att(x)
     """
 
+    @typechecked
     def __init__(self,
                  vocab_size: int,
                  encoder_output_size: int,
@@ -293,8 +293,6 @@ class BiTransformerDecoder(BatchScorerInterface, nn.Layer):
                  concat_after: bool=False,
                  max_len: int=5000):
 
-        assert check_argument_types()
-
         nn.Layer.__init__(self)
         self.left_decoder = TransformerDecoder(
             vocab_size, encoder_output_size, attention_heads, linear_units,

paddlespeech/s2t/modules/encoder.py

@@ -21,7 +21,7 @@ from typing import Union
 
 import paddle
 from paddle import nn
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 from paddlespeech.s2t.modules.activation import get_activation
 from paddlespeech.s2t.modules.align import LayerNorm
@@ -58,6 +58,7 @@ __all__ = [
 
 
 class BaseEncoder(nn.Layer):
+    @typechecked
     def __init__(self,
                  input_size: int,
                  output_size: int=256,
@@ -73,7 +74,7 @@ class BaseEncoder(nn.Layer):
                  concat_after: bool=False,
                  static_chunk_size: int=0,
                  use_dynamic_chunk: bool=False,
-                 global_cmvn: paddle.nn.Layer=None,
+                 global_cmvn: Optional[nn.Layer]=None,
                  use_dynamic_left_chunk: bool=False,
                  max_len: int=5000):
         """
@@ -108,7 +109,6 @@ class BaseEncoder(nn.Layer):
             use_dynamic_left_chunk (bool): whether use dynamic left chunk in
                 dynamic chunk training
         """
-        assert check_argument_types()
         super().__init__()
         self._output_size = output_size
 
@@ -349,6 +349,7 @@ class BaseEncoder(nn.Layer):
 class TransformerEncoder(BaseEncoder):
     """Transformer encoder module."""
 
+    @typechecked
     def __init__(
             self,
             input_size: int,
@@ -365,12 +366,11 @@ class TransformerEncoder(BaseEncoder):
             concat_after: bool=False,
             static_chunk_size: int=0,
             use_dynamic_chunk: bool=False,
-            global_cmvn: nn.Layer=None,
+            global_cmvn: Optional[nn.Layer]=None,
             use_dynamic_left_chunk: bool=False, ):
         """ Construct TransformerEncoder
         See Encoder for the meaning of each parameter.
         """
-        assert check_argument_types()
         super().__init__(input_size, output_size, attention_heads, linear_units,
                          num_blocks, dropout_rate, positional_dropout_rate,
                          attention_dropout_rate, input_layer,
@@ -424,6 +424,7 @@ class TransformerEncoder(BaseEncoder):
 class ConformerEncoder(BaseEncoder):
     """Conformer encoder module."""
 
+    @typechecked
     def __init__(self,
                  input_size: int,
                  output_size: int=256,
@@ -439,7 +440,7 @@ class ConformerEncoder(BaseEncoder):
                  concat_after: bool=False,
                  static_chunk_size: int=0,
                  use_dynamic_chunk: bool=False,
-                 global_cmvn: nn.Layer=None,
+                 global_cmvn: Optional[nn.Layer]=None,
                  use_dynamic_left_chunk: bool=False,
                  positionwise_conv_kernel_size: int=1,
                  macaron_style: bool=True,
@@ -466,8 +467,6 @@ class ConformerEncoder(BaseEncoder):
             causal (bool): whether to use causal convolution or not.
             cnn_module_norm (str): cnn conv norm type, Optional['batch_norm','layer_norm']
         """
-        assert check_argument_types()
-
         super().__init__(input_size, output_size, attention_heads, linear_units,
                          num_blocks, dropout_rate, positional_dropout_rate,
                          attention_dropout_rate, input_layer,
@@ -519,6 +518,7 @@ class ConformerEncoder(BaseEncoder):
 
 
 class SqueezeformerEncoder(nn.Layer):
+    @typechecked
     def __init__(self,
                  input_size: int,
                  encoder_dim: int=256,
@@ -541,7 +541,7 @@ class SqueezeformerEncoder(nn.Layer):
                  adaptive_scale: bool=True,
                  activation_type: str="swish",
                  init_weights: bool=True,
-                 global_cmvn: paddle.nn.Layer=None,
+                 global_cmvn: Optional[nn.Layer]=None,
                  normalize_before: bool=False,
                  use_dynamic_chunk: bool=False,
                  concat_after: bool=False,
@@ -572,7 +572,6 @@ class SqueezeformerEncoder(nn.Layer):
             init_weights (bool): Whether to initialize weights.
             causal (bool): whether to use causal convolution or not.
         """
-        assert check_argument_types()
         super().__init__()
         self.global_cmvn = global_cmvn
         self.reduce_idx: Optional[Union[int, List[int]]] = [reduce_idx] \

paddlespeech/s2t/training/scheduler.py

@@ -19,7 +19,7 @@ from typing import Union
 
 import paddle
 from paddle.optimizer.lr import LRScheduler
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 from paddlespeech.s2t.utils.dynamic_import import dynamic_import
 from paddlespeech.s2t.utils.dynamic_import import instance_class
@@ -57,13 +57,13 @@ class WarmupLR(LRScheduler):
     Note that the maximum lr equals to optimizer.lr in this scheduler.
     """
 
+    @typechecked
     def __init__(self,
                  warmup_steps: Union[int, float]=25000,
                  learning_rate=1.0,
                  last_epoch=-1,
                  verbose=False,
                  **kwargs):
-        assert check_argument_types()
         self.warmup_steps = warmup_steps
         super().__init__(learning_rate, last_epoch, verbose)
 

paddlespeech/t2s/models/diffsinger/diffsinger.py

@@ -20,7 +20,7 @@ from typing import Tuple
 import numpy as np
 import paddle
 from paddle import nn
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 from paddlespeech.t2s.models.diffsinger.fastspeech2midi import FastSpeech2MIDI
 from paddlespeech.t2s.modules.diffnet import DiffNet
@@ -40,6 +40,7 @@ class DiffSinger(nn.Layer):
 
     """
 
+    @typechecked
     def __init__(
             self,
             # min and max spec for stretching before diffusion
@@ -157,7 +158,6 @@ class DiffSinger(nn.Layer):
             denoiser_params (Dict[str, Any]): Parameter dict for dinoiser module.
             diffusion_params (Dict[str, Any]): Parameter dict for diffusion module.
         """
-        assert check_argument_types()
         super().__init__()
         self.fs2 = FastSpeech2MIDI(
             idim=idim,
@@ -336,6 +336,7 @@ class DiffSingerInference(nn.Layer):
 class DiffusionLoss(nn.Layer):
     """Loss function module for Diffusion module on DiffSinger."""
 
+    @typechecked
     def __init__(self, use_masking: bool=True,
                  use_weighted_masking: bool=False):
         """Initialize feed-forward Transformer loss module.
@@ -345,7 +346,6 @@ class DiffusionLoss(nn.Layer):
             use_weighted_masking (bool): 
                 Whether to weighted masking in loss calculation.
         """
-        assert check_argument_types()
         super().__init__()
 
         assert (use_masking != use_weighted_masking) or not use_masking

paddlespeech/t2s/models/diffsinger/fastspeech2midi.py

@@ -19,7 +19,7 @@ from typing import Tuple
 
 import paddle
 from paddle import nn
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 from paddlespeech.t2s.models.fastspeech2 import FastSpeech2
 from paddlespeech.t2s.models.fastspeech2 import FastSpeech2Loss
@@ -33,6 +33,7 @@ class FastSpeech2MIDI(FastSpeech2):
     """The Fastspeech2 module of DiffSinger.
     """
 
+    @typechecked
     def __init__(
             self,
             # fastspeech2 network structure related
@@ -57,7 +58,6 @@ class FastSpeech2MIDI(FastSpeech2):
                 is_slur_ids will be provided as the input
     
         """
-        assert check_argument_types()
         super().__init__(idim=idim, odim=odim, **fastspeech2_params)
         self.use_energy_pred = use_energy_pred
         self.use_postnet = use_postnet
@@ -495,6 +495,7 @@ class FastSpeech2MIDI(FastSpeech2):
 class FastSpeech2MIDILoss(FastSpeech2Loss):
     """Loss function module for DiffSinger."""
 
+    @typechecked
     def __init__(self, use_masking: bool=True,
                  use_weighted_masking: bool=False):
         """Initialize feed-forward Transformer loss module.
@@ -504,7 +505,6 @@ class FastSpeech2MIDILoss(FastSpeech2Loss):
             use_weighted_masking (bool): 
                 Whether to weighted masking in loss calculation.
         """
-        assert check_argument_types()
         super().__init__(use_masking, use_weighted_masking)
 
     def forward(

paddlespeech/t2s/models/fastspeech2/fastspeech2.py

@@ -15,6 +15,7 @@
 """Fastspeech2 related modules for paddle"""
 from typing import Dict
 from typing import List
+from typing import Optional
 from typing import Sequence
 from typing import Tuple
 from typing import Union
@@ -23,7 +24,7 @@ import numpy as np
 import paddle
 import paddle.nn.functional as F
 from paddle import nn
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 from paddlespeech.t2s.modules.adversarial_loss.gradient_reversal import GradientReversalLayer
 from paddlespeech.t2s.modules.adversarial_loss.speaker_classifier import SpeakerClassifier
@@ -60,6 +61,7 @@ class FastSpeech2(nn.Layer):
 
     """
 
+    @typechecked
     def __init__(
             self,
             # network structure related
@@ -131,12 +133,12 @@ class FastSpeech2(nn.Layer):
             pitch_embed_dropout: float=0.5,
             stop_gradient_from_pitch_predictor: bool=False,
             # spk emb
-            spk_num: int=None,
+            spk_num: Optional[int]=None,
-            spk_embed_dim: int=None,
+            spk_embed_dim: Optional[int]=None,
             spk_embed_integration_type: str="add",
             # tone emb
-            tone_num: int=None,
+            tone_num: Optional[int]=None,
-            tone_embed_dim: int=None,
+            tone_embed_dim: Optional[int]=None,
             tone_embed_integration_type: str="add",
             # training related
             init_type: str="xavier_uniform",
@@ -282,7 +284,6 @@ class FastSpeech2(nn.Layer):
                 The hidden layer dim of speaker classifier
     
         """
-        assert check_argument_types()
         super().__init__()
 
         # store hyperparameters
@@ -1070,6 +1071,7 @@ class StyleFastSpeech2Inference(FastSpeech2Inference):
 class FastSpeech2Loss(nn.Layer):
     """Loss function module for FastSpeech2."""
 
+    @typechecked
     def __init__(self, use_masking: bool=True,
                  use_weighted_masking: bool=False):
         """Initialize feed-forward Transformer loss module.
@@ -1079,7 +1081,6 @@ class FastSpeech2Loss(nn.Layer):
             use_weighted_masking (bool): 
                 Whether to weighted masking in loss calculation.
         """
-        assert check_argument_types()
         super().__init__()
 
         assert (use_masking != use_weighted_masking) or not use_masking

paddlespeech/t2s/models/jets/generator.py

@@ -28,7 +28,6 @@ from typing import Tuple
 import numpy as np
 import paddle
 from paddle import nn
-from typeguard import check_argument_types
 
 from paddlespeech.t2s.models.hifigan import HiFiGANGenerator
 from paddlespeech.t2s.models.jets.alignments import AlignmentModule

paddlespeech/t2s/models/jets/jets.py

@@ -24,7 +24,7 @@ from typing import Optional
 
 import paddle
 from paddle import nn
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 from paddlespeech.t2s.models.hifigan import HiFiGANMultiPeriodDiscriminator
 from paddlespeech.t2s.models.hifigan import HiFiGANMultiScaleDiscriminator
@@ -64,6 +64,7 @@ class JETS(nn.Layer):
         Text-to-Speech`: https://arxiv.org/abs/2203.16852v1
     """
 
+    @typechecked
     def __init__(
             self,
             # generator related
@@ -225,7 +226,6 @@ class JETS(nn.Layer):
             cache_generator_outputs (bool):
                 Whether to cache generator outputs.
         """
-        assert check_argument_types()
         super().__init__()
 
         # define modules
@@ -279,8 +279,7 @@ class JETS(nn.Layer):
             lids: Optional[paddle.Tensor]=None,
             forward_generator: bool=True,
             use_alignment_module: bool=False,
-            **kwargs,
+            **kwargs, ) -> Dict[str, Any]:
-    ) -> Dict[str, Any]:
         """Perform generator forward.
         Args:
             text (Tensor):

paddlespeech/t2s/models/tacotron2/tacotron2.py

@@ -21,7 +21,7 @@ from typing import Tuple
 import paddle
 import paddle.nn.functional as F
 from paddle import nn
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 from paddlespeech.t2s.modules.nets_utils import initialize
 from paddlespeech.t2s.modules.nets_utils import make_pad_mask
@@ -44,6 +44,7 @@ class Tacotron2(nn.Layer):
 
     """
 
+    @typechecked
     def __init__(
             self,
             # network structure related
@@ -67,7 +68,7 @@ class Tacotron2(nn.Layer):
             postnet_layers: int=5,
             postnet_chans: int=512,
             postnet_filts: int=5,
-            output_activation: str=None,
+            output_activation: Optional[str]=None,
             use_batch_norm: bool=True,
             use_concate: bool=True,
             use_residual: bool=False,
@@ -145,7 +146,6 @@ class Tacotron2(nn.Layer):
             zoneout_rate (float): 
                 Zoneout rate.
         """
-        assert check_argument_types()
         super().__init__()
 
         # store hyperparameters

paddlespeech/t2s/models/transformer_tts/transformer_tts.py

@@ -13,7 +13,9 @@
 # limitations under the License.
 # Modified from espnet(https://github.com/espnet/espnet)
 """Fastspeech2 related modules for paddle"""
+from optparse import Option
 from typing import Dict
+from typing import Optional
 from typing import Sequence
 from typing import Tuple
 
@@ -21,7 +23,7 @@ import numpy
 import paddle
 import paddle.nn.functional as F
 from paddle import nn
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 from paddlespeech.t2s.modules.nets_utils import initialize
 from paddlespeech.t2s.modules.nets_utils import make_non_pad_mask
@@ -169,6 +171,7 @@ class TransformerTTS(nn.Layer):
             Number of layers to apply guided attention loss.
     """
 
+    @typechecked
     def __init__(
             self,
             # network structure related
@@ -198,7 +201,7 @@ class TransformerTTS(nn.Layer):
             encoder_concat_after: bool=False,
             decoder_concat_after: bool=False,
             reduction_factor: int=1,
-            spk_embed_dim: int=None,
+            spk_embed_dim: Optional[int]=None,
             spk_embed_integration_type: str="add",
             use_gst: bool=False,
             gst_tokens: int=10,
@@ -227,7 +230,7 @@ class TransformerTTS(nn.Layer):
             num_heads_applied_guided_attn: int=2,
             num_layers_applied_guided_attn: int=2, ):
         """Initialize Transformer module."""
-        assert check_argument_types()
+
         super().__init__()
 
         # store hyperparameters

paddlespeech/t2s/models/vits/vits.py

@@ -20,7 +20,7 @@ from typing import Optional
 
 import paddle
 from paddle import nn
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 from paddlespeech.t2s.models.hifigan import HiFiGANMultiPeriodDiscriminator
 from paddlespeech.t2s.models.hifigan import HiFiGANMultiScaleDiscriminator
@@ -60,6 +60,7 @@ class VITS(nn.Layer):
         Text-to-Speech`: https://arxiv.org/abs/2006.04558
     """
 
+    @typechecked
     def __init__(
             self,
             # generator related
@@ -181,7 +182,6 @@ class VITS(nn.Layer):
             cache_generator_outputs (bool):
                 Whether to cache generator outputs.
         """
-        assert check_argument_types()
         super().__init__()
 
         # define modules
@@ -504,8 +504,9 @@ class VITS(nn.Layer):
 
     def reset_parameters(self):
         def _reset_parameters(module):
-            if isinstance(module,
+            if isinstance(
-                        (nn.Conv1D, nn.Conv1DTranspose, nn.Conv2D, nn.Conv2DTranspose)):
+                    module,
+                (nn.Conv1D, nn.Conv1DTranspose, nn.Conv2D, nn.Conv2DTranspose)):
                 kaiming_uniform_(module.weight, a=math.sqrt(5))
                 if module.bias is not None:
                     fan_in, _ = _calculate_fan_in_and_fan_out(module.weight)
@@ -513,8 +514,9 @@ class VITS(nn.Layer):
                         bound = 1 / math.sqrt(fan_in)
                         uniform_(module.bias, -bound, bound)
 
-            if isinstance(module,
+            if isinstance(
-                          (nn.BatchNorm1D, nn.BatchNorm2D, nn.GroupNorm, nn.LayerNorm)):
+                    module,
+                (nn.BatchNorm1D, nn.BatchNorm2D, nn.GroupNorm, nn.LayerNorm)):
                 ones_(module.weight)
                 zeros_(module.bias)
 
@@ -533,13 +535,13 @@ class VITS(nn.Layer):
 
         self.apply(_reset_parameters)
 
+
 class VITSInference(nn.Layer):
     def __init__(self, model):
         super().__init__()
         self.acoustic_model = model
 
     def forward(self, text, sids=None):
-        out = self.acoustic_model.inference(
+        out = self.acoustic_model.inference(text, sids=sids)
-            text, sids=sids)
         wav = out['wav']
         return wav

paddlespeech/t2s/modules/adversarial_loss/speaker_classifier.py

@@ -14,16 +14,16 @@
 # Modified from Cross-Lingual-Voice-Cloning(https://github.com/deterministic-algorithms-lab/Cross-Lingual-Voice-Cloning)
 import paddle
 from paddle import nn
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 
 class SpeakerClassifier(nn.Layer):
+    @typechecked
     def __init__(
             self,
             idim: int,
             hidden_sc_dim: int,
             spk_num: int, ):
-        assert check_argument_types()
         super().__init__()
         # store hyperparameters
         self.idim = idim

paddlespeech/t2s/modules/losses.py

@@ -21,7 +21,7 @@ from paddle import nn
 from paddle.nn import functional as F
 from scipy import signal
 from scipy.stats import betabinom
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 from paddlespeech.t2s.modules.nets_utils import make_non_pad_mask
 from paddlespeech.t2s.modules.predictor.duration_predictor import (
@@ -1137,6 +1137,7 @@ class MLMLoss(nn.Layer):
 
 
 class VarianceLoss(nn.Layer):
+    @typechecked
     def __init__(self, use_masking: bool=True,
                  use_weighted_masking: bool=False):
         """Initialize JETS variance loss module.
@@ -1147,7 +1148,6 @@ class VarianceLoss(nn.Layer):
                 calculation.
 
         """
-        assert check_argument_types()
         super().__init__()
 
         assert (use_masking != use_weighted_masking) or not use_masking

paddlespeech/t2s/modules/nets_utils.py

@@ -18,7 +18,7 @@ from typing import Tuple
 import numpy as np
 import paddle
 from paddle import nn
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 from paddlespeech.utils.initialize import _calculate_fan_in_and_fan_out
 from paddlespeech.utils.initialize import kaiming_uniform_
@@ -301,6 +301,7 @@ def make_non_pad_mask(lengths, xs=None, length_dim=-1):
     return paddle.logical_not(make_pad_mask(lengths, xs, length_dim))
 
 
+@typechecked
 def initialize(model: nn.Layer, init: str):
     """Initialize weights of a neural network module.
 
@@ -314,8 +315,6 @@ def initialize(model: nn.Layer, init: str):
         init (str):
             Method of initialization.
     """
-    assert check_argument_types()
-
     if init == "xavier_uniform":
         nn.initializer.set_global_initializer(nn.initializer.XavierUniform(),
                                               nn.initializer.Constant())

paddlespeech/t2s/modules/predictor/variance_predictor.py

@@ -15,7 +15,7 @@
 """Variance predictor related modules."""
 import paddle
 from paddle import nn
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 from paddlespeech.t2s.modules.layer_norm import LayerNorm
 from paddlespeech.t2s.modules.masked_fill import masked_fill
@@ -32,6 +32,7 @@ class VariancePredictor(nn.Layer):
 
     """
 
+    @typechecked
     def __init__(
             self,
             idim: int,
@@ -54,7 +55,6 @@ class VariancePredictor(nn.Layer):
             dropout_rate (float, optional): 
                 Dropout rate.
         """
-        assert check_argument_types()
         super().__init__()
         self.conv = nn.LayerList()
         for idx in range(n_layers):
@@ -96,7 +96,7 @@ class VariancePredictor(nn.Layer):
             xs = f(xs)
         # (B, Tmax, 1)
         xs = self.linear(xs.transpose([0, 2, 1]))
-    
+
         if x_masks is not None:
             xs = masked_fill(xs, x_masks, 0.0)
         return xs

paddlespeech/t2s/modules/style_encoder.py

@@ -17,7 +17,7 @@ from typing import Sequence
 
 import paddle
 from paddle import nn
-from typeguard import check_argument_types
+from typeguard import typechecked
 
 from paddlespeech.t2s.modules.transformer.attention import MultiHeadedAttention as BaseMultiHeadedAttention
 
@@ -58,6 +58,7 @@ class StyleEncoder(nn.Layer):
 
     """
 
+    @typechecked
     def __init__(
             self,
             idim: int=80,
@@ -71,7 +72,6 @@ class StyleEncoder(nn.Layer):
             gru_layers: int=1,
             gru_units: int=128, ):
         """Initilize global style encoder module."""
-        assert check_argument_types()
         super().__init__()
 
         self.ref_enc = ReferenceEncoder(
@@ -132,6 +132,7 @@ class ReferenceEncoder(nn.Layer):
 
     """
 
+    @typechecked
     def __init__(
             self,
             idim=80,
@@ -142,7 +143,6 @@ class ReferenceEncoder(nn.Layer):
             gru_layers: int=1,
             gru_units: int=128, ):
         """Initilize reference encoder module."""
-        assert check_argument_types()
         super().__init__()
 
         # check hyperparameters are valid
@@ -232,6 +232,7 @@ class StyleTokenLayer(nn.Layer):
 
     """
 
+    @typechecked
     def __init__(
             self,
             ref_embed_dim: int=128,
@@ -240,7 +241,6 @@ class StyleTokenLayer(nn.Layer):
             gst_heads: int=4,
             dropout_rate: float=0.0, ):
         """Initilize style token layer module."""
-        assert check_argument_types()
         super().__init__()
 
         gst_embs = paddle.randn(shape=[gst_tokens, gst_token_dim // gst_heads])

setup.py

@@ -31,6 +31,26 @@ HERE = Path(os.path.abspath(os.path.dirname(__file__)))
 VERSION = '0.0.0'
 COMMITID = 'none'
 
+
+def determine_opencc_version():
+    # get gcc version
+    gcc_version = None
+    try:
+        output = sp.check_output(
+            ['gcc', '--version'], stderr=sp.STDOUT, text=True)
+        for line in output.splitlines():
+            if "gcc" in line:
+                gcc_version = line.split()[-1]
+    except Exception as e:
+        gcc_version = None
+
+    # determine opencc version
+    if gcc_version:
+        if int(gcc_version.split(".")[0]) <= 9:
+            return "opencc==1.1.6"  # GCC<=9 need opencc==1.1.6
+    return "opencc"  # default
+
+
 base = [
     "braceexpand",
     "editdistance",
@@ -48,7 +68,7 @@ base = [
     "matplotlib",
     "nara_wpe",
     "onnxruntime>=1.11.0",
-    "opencc==1.1.6",
+    determine_opencc_version(),  # opencc or opencc==1.1.6
     "opencc-python-reimplemented",
     "pandas",
     "paddleaudio>=1.1.0",
@@ -69,8 +89,8 @@ base = [
     "soundfile",
     "textgrid",
     "timer",
-    "ToJyutping==0.2.1",
+    "ToJyutping",
-    "typeguard==2.13.3",
+    "typeguard",
     "webrtcvad",
     "yacs~=0.1.8",
     "zhon",
@@ -318,9 +338,9 @@ setup_info = dict(
         'License :: OSI Approved :: Apache Software License',
         'Programming Language :: Python',
         'Programming Language :: Python :: 3',
-        'Programming Language :: Python :: 3.7',
         'Programming Language :: Python :: 3.8',
         'Programming Language :: Python :: 3.9',
+        'Programming Language :: Python :: 3.10',
     ],
     entry_points={
         'console_scripts': [

tests/unit/audiotools/core/test_audio_signal.py

@@ -12,9 +12,9 @@ import paddle
 import pytest
 import rich
 
-from audio import audiotools
+from paddlespeech import audiotools
-from audio.audiotools import AudioSignal
+from paddlespeech.audiotools import AudioSignal
-from audio.audiotools import util
+from paddlespeech.audiotools import util
 
 
 def test_io():

tests/unit/audiotools/core/test_bands.py

@@ -8,9 +8,9 @@ import unittest
 
 import paddle
 
-from audio.audiotools.core import pure_tone
+from paddlespeech.audiotools.core import pure_tone
-from audio.audiotools.core import split_bands
+from paddlespeech.audiotools.core import split_bands
-from audio.audiotools.core import SplitBands
+from paddlespeech.audiotools.core import SplitBands
 
 
 def delta(a, b, ref, fraction=0.9):

tests/unit/audiotools/core/test_display.py

@@ -8,7 +8,7 @@ from pathlib import Path
 import numpy as np
 from visualdl import LogWriter
 
-from audio.audiotools import AudioSignal
+from paddlespeech.audiotools import AudioSignal
 
 
 def test_specshow():

tests/unit/audiotools/core/test_dsp.py

@@ -8,8 +8,8 @@ import numpy as np
 import paddle
 import pytest
 
-from audio.audiotools import AudioSignal
+from paddlespeech.audiotools import AudioSignal
-from audio.audiotools.core.util import sample_from_dist
+from paddlespeech.audiotools.core.util import sample_from_dist
 
 
 @pytest.mark.parametrize("window_duration", [0.1, 0.25, 0.5, 1.0])

tests/unit/audiotools/core/test_effects.py

@@ -8,7 +8,7 @@ import numpy as np
 import paddle
 import pytest
 
-from audio.audiotools import AudioSignal
+from paddlespeech.audiotools import AudioSignal
 
 
 def test_normalize():

tests/unit/audiotools/core/test_fftconv.py

@@ -9,8 +9,8 @@ import unittest
 import paddle
 import paddle.nn.functional as F
 
-from audio.audiotools.core import fft_conv1d
+from paddlespeech.audiotools.core import fft_conv1d
-from audio.audiotools.core import FFTConv1D
+from paddlespeech.audiotools.core import FFTConv1D
 
 TOLERANCE = 1e-4  # as relative delta in percentage
 

tests/unit/audiotools/core/test_grad.py

@@ -9,7 +9,7 @@ import numpy as np
 import paddle
 import pytest
 
-from audio.audiotools import AudioSignal
+from paddlespeech.audiotools import AudioSignal
 
 
 def test_audio_grad():

tests/unit/audiotools/core/test_highpass.py

@@ -9,8 +9,8 @@ import unittest
 
 import paddle
 
-from audio.audiotools.core import highpass_filter
+from paddlespeech.audiotools.core import highpass_filter
-from audio.audiotools.core import highpass_filters
+from paddlespeech.audiotools.core import highpass_filters
 
 
 def pure_tone(freq: float, sr: float=128, dur: float=4, device=None):

tests/unit/audiotools/core/test_loudness.py

@@ -8,10 +8,10 @@ import numpy as np
 import pyloudnorm
 import soundfile as sf
 
-from audio.audiotools import AudioSignal
+from paddlespeech.audiotools import AudioSignal
-from audio.audiotools import datasets
+from paddlespeech.audiotools import datasets
-from audio.audiotools import Meter
+from paddlespeech.audiotools import Meter
-from audio.audiotools import transforms
+from paddlespeech.audiotools import transforms
 
 ATOL = 1e-1
 

tests/unit/audiotools/core/test_lowpass.py

@@ -10,10 +10,10 @@ import unittest
 import numpy as np
 import paddle
 
-from audio.audiotools.core import lowpass_filter
+from paddlespeech.audiotools.core import lowpass_filter
-from audio.audiotools.core import LowPassFilter
+from paddlespeech.audiotools.core import LowPassFilter
-from audio.audiotools.core import LowPassFilters
+from paddlespeech.audiotools.core import LowPassFilters
-from audio.audiotools.core import resample_frac
+from paddlespeech.audiotools.core import resample_frac
 
 
 def pure_tone(freq: float, sr: float=128, dur: float=4, device=None):

tests/unit/audiotools/core/test_util.py

@@ -11,8 +11,8 @@ import numpy as np
 import paddle
 import pytest
 
-from audio.audiotools import util
+from paddlespeech.audiotools import util
-from audio.audiotools.core.audio_signal import AudioSignal
+from paddlespeech.audiotools.core.audio_signal import AudioSignal
 from paddlespeech.vector.training.seeding import seed_everything
 
 

tests/unit/audiotools/data/test_datasets.py

@@ -10,8 +10,8 @@ import numpy as np
 import paddle
 import pytest
 
-from audio import audiotools
+from paddlespeech import audiotools
-from audio.audiotools.data import transforms as tfm
+from paddlespeech.audiotools.data import transforms as tfm
 
 
 def test_align_lists():

tests/unit/audiotools/data/test_preprocess.py

@@ -8,9 +8,9 @@ from pathlib import Path
 
 import paddle
 
-from audio.audiotools.core.util import find_audio
+from paddlespeech.audiotools.core.util import find_audio
-from audio.audiotools.core.util import read_sources
+from paddlespeech.audiotools.core.util import read_sources
-from audio.audiotools.data import preprocess
+from paddlespeech.audiotools.data import preprocess
 
 
 def test_create_csv():

tests/unit/audiotools/data/test_transforms.py

@@ -11,11 +11,11 @@ import numpy as np
 import paddle
 import pytest
 
-from audio import audiotools
+from paddlespeech import audiotools
-from audio.audiotools import AudioSignal
+from paddlespeech.audiotools import AudioSignal
-from audio.audiotools import util
+from paddlespeech.audiotools import util
-from audio.audiotools.data import transforms as tfm
+from paddlespeech.audiotools.data import transforms as tfm
-from audio.audiotools.data.datasets import AudioDataset
+from paddlespeech.audiotools.data.datasets import AudioDataset
 from paddlespeech.vector.training.seeding import seed_everything
 
 non_deterministic_transforms = ["TimeNoise", "FrequencyNoise"]

tests/unit/audiotools/ml/test_decorators.py

@@ -8,10 +8,10 @@ import time
 import paddle
 from visualdl import LogWriter
 
-from audio.audiotools import util
+from paddlespeech.audiotools import util
-from audio.audiotools.ml.decorators import timer
+from paddlespeech.audiotools.ml.decorators import timer
-from audio.audiotools.ml.decorators import Tracker
+from paddlespeech.audiotools.ml.decorators import Tracker
-from audio.audiotools.ml.decorators import when
+from paddlespeech.audiotools.ml.decorators import when
 
 
 def test_all_decorators():

tests/unit/audiotools/ml/test_model.py

@@ -8,10 +8,10 @@ import tempfile
 import paddle
 from paddle import nn
 
-from audio.audiotools import ml
+from paddlespeech.audiotools import ml
-from audio.audiotools import util
+from paddlespeech.audiotools import util
 from paddlespeech.vector.training.seeding import seed_everything
-SEED = 0
+SEED = 1024
 
 
 def seed_and_run(model, *args, **kwargs):

tests/unit/audiotools/test_audiotools.sh

@@ -1,5 +1,4 @@
-python -m pip install -r ../../audiotools/requirements.txt
+python -m pip install -r ../../../paddlespeech/audiotools/requirements.txt
-export PYTHONPATH=$PYTHONPATH:$(realpath ../../..) # this is root path of `PaddleSpeech`
 wget  https://paddlespeech.bj.bcebos.com/PaddleAudio/audio_tools/audio.tar.gz
 wget  https://paddlespeech.bj.bcebos.com/PaddleAudio/audio_tools/regression.tar.gz
 tar -zxvf audio.tar.gz

tests/unit/audiotools/test_post.py

@@ -5,9 +5,9 @@
 import sys
 from pathlib import Path
 
-from audio.audiotools import AudioSignal
+from paddlespeech.audiotools import AudioSignal
-from audio.audiotools import post
+from paddlespeech.audiotools import post
-from audio.audiotools import transforms
+from paddlespeech.audiotools import transforms
 
 
 def test_audio_table():

tests/unit/ci.sh

@@ -34,7 +34,7 @@ function main(){
   echo "End server"
 
   echo "Start testing audiotools"
-  cd ${speech_ci_path}/../../audio/tests/audiotools
+  cd ${speech_ci_path}/audiotools
   bash test_audiotools.sh
   echo "End testing audiotools"
 

tests/unit/tts/test_expansion.py

@@ -13,10 +13,10 @@
 # limitations under the License.
 import paddle
 
-from paddlespeech.t2s.modules import expansion
+# from paddlespeech.t2s.modules import expansion
 
 
-def test_expand():
+def _test_expand():
     x = paddle.randn([2, 4, 3])  # (B, T, C)
     lengths = paddle.to_tensor([[1, 2, 2, 1], [3, 1, 4, 0]])
     y = expansion.expand(x, lengths)