v0.8.0

游雁

2023-10-10 580b11b57ac4b62f7e2acda73813a4e10e8e4cd3

 funasr/tasks/diar.py

@@ -13,17 +13,15 @@
import numpy as np
import torch
import yaml
from typeguard import check_argument_types
from typeguard import check_return_type

from funasr.datasets.collate_fn import CommonCollateFn
from funasr.datasets.collate_fn import DiarCollateFn
from funasr.datasets.preprocessor import CommonPreprocessor
from funasr.layers.abs_normalize import AbsNormalize
from funasr.layers.global_mvn import GlobalMVN
from funasr.layers.utterance_mvn import UtteranceMVN
from funasr.layers.label_aggregation import LabelAggregate
from funasr.layers.label_aggregation import LabelAggregate, LabelAggregateMaxPooling
from funasr.models.ctc import CTC
from funasr.models.encoder.resnet34_encoder import ResNet34Diar
from funasr.models.encoder.resnet34_encoder import ResNet34Diar, ResNet34SpL2RegDiar
from funasr.models.encoder.ecapa_tdnn_encoder import ECAPA_TDNN
from funasr.models.encoder.opennmt_encoders.conv_encoder import ConvEncoder
from funasr.models.encoder.opennmt_encoders.fsmn_encoder import FsmnEncoder
@@ -52,9 +50,10 @@
from funasr.models.specaug.abs_specaug import AbsSpecAug
from funasr.models.specaug.specaug import SpecAug
from funasr.models.specaug.specaug import SpecAugLFR
from funasr.models.specaug.abs_profileaug import AbsProfileAug
from funasr.models.specaug.profileaug import ProfileAug
from funasr.tasks.abs_task import AbsTask
from funasr.torch_utils.initialize import initialize
from funasr.train.abs_espnet_model import AbsESPnetModel
from funasr.train.class_choices import ClassChoices
from funasr.train.trainer import Trainer
from funasr.utils.types import float_or_none
@@ -84,6 +83,15 @@
    default=None,
    optional=True,
)
profileaug_choices = ClassChoices(
    name="profileaug",
    classes=dict(
        profileaug=ProfileAug,
    ),
    type_check=AbsProfileAug,
    default=None,
    optional=True,
)
normalize_choices = ClassChoices(
    "normalize",
    classes=dict(
@@ -97,7 +105,8 @@
label_aggregator_choices = ClassChoices(
    "label_aggregator",
    classes=dict(
        label_aggregator=LabelAggregate
        label_aggregator=LabelAggregate,
        label_aggregator_max_pool=LabelAggregateMaxPooling,
    ),
    type_check=torch.nn.Module,
    default=None,
@@ -108,7 +117,7 @@
    classes=dict(
        sond=DiarSondModel,
    ),
    type_check=AbsESPnetModel,
    type_check=torch.nn.Module,
    default="sond",
)
encoder_choices = ClassChoices(
@@ -122,6 +131,7 @@
        fsmn=FsmnEncoder,
        conv=ConvEncoder,
        resnet34=ResNet34Diar,
        resnet34_sp_l2reg=ResNet34SpL2RegDiar,
        sanm_chunk_opt=SANMEncoderChunkOpt,
        data2vec_encoder=Data2VecEncoder,
        ecapa_tdnn=ECAPA_TDNN,
@@ -160,6 +170,7 @@
    classes=dict(
        dot=DotScorer,
        cosine=CosScorer,
        conv=ConvEncoder,
    ),
    type_check=torch.nn.Module,
    default=None,
@@ -187,6 +198,8 @@
        frontend_choices,
        # --specaug and --specaug_conf
        specaug_choices,
        # --profileaug and --profileaug_conf
        profileaug_choices,
        # --normalize and --normalize_conf
        normalize_choices,
        # --label_aggregator and --label_aggregator_conf
@@ -326,15 +339,13 @@
        [Collection[Tuple[str, Dict[str, np.ndarray]]]],
        Tuple[List[str], Dict[str, torch.Tensor]],
    ]:
        assert check_argument_types()
        # NOTE(kamo): int value = 0 is reserved by CTC-blank symbol
        return CommonCollateFn(float_pad_value=0.0, int_pad_value=-1)
        return DiarCollateFn(float_pad_value=0.0, int_pad_value=-1)

    @classmethod
    def build_preprocess_fn(
            cls, args: argparse.Namespace, train: bool
    ) -> Optional[Callable[[str, Dict[str, np.array]], Dict[str, np.ndarray]]]:
        assert check_argument_types()
        if args.use_preprocessor:
            retval = CommonPreprocessor(
                train=train,
@@ -364,7 +375,6 @@
            )
        else:
            retval = None
        assert check_return_type(retval)
        return retval

    @classmethod
@@ -383,12 +393,45 @@
            cls, train: bool = True, inference: bool = False
    ) -> Tuple[str, ...]:
        retval = ()
        assert check_return_type(retval)
        return retval

    @classmethod
    def build_optimizers(
            cls,
            args: argparse.Namespace,
            model: torch.nn.Module,
    ) -> List[torch.optim.Optimizer]:
        if cls.num_optimizers != 1:
            raise RuntimeError(
                "build_optimizers() must be overridden if num_optimizers != 1"
            )
        from funasr.tasks.abs_task import optim_classes
        optim_class = optim_classes.get(args.optim)
        if optim_class is None:
            raise ValueError(f"must be one of {list(optim_classes)}: {args.optim}")
        else:
            if (hasattr(model, "model_regularizer_weight") and
                model.model_regularizer_weight > 0.0 and
                hasattr(model, "get_regularize_parameters")
            ):
                to_regularize_parameters, normal_parameters = model.get_regularize_parameters()
                logging.info(f"Set weight decay {model.model_regularizer_weight} for parameters: "
                             f"{[name for name, value in to_regularize_parameters]}")
                module_optim_config = [
                    {"params": [value for name, value in to_regularize_parameters],
                     "weight_decay": model.model_regularizer_weight},
                    {"params": [value for name, value in normal_parameters],
                     "weight_decay": 0.0}
                ]
                optim = optim_class(module_optim_config, **args.optim_conf)
            else:
                optim = optim_class(model.parameters(), **args.optim_conf)

        optimizers = [optim]
        return optimizers

    @classmethod
    def build_model(cls, args: argparse.Namespace):
        assert check_argument_types()
        if isinstance(args.token_list, str):
            with open(args.token_list, encoding="utf-8") as f:
                token_list = [line.rstrip() for line in f]
@@ -424,6 +467,13 @@
            specaug = specaug_class(**args.specaug_conf)
        else:
            specaug = None

        # 2b. Data augmentation for Profiles
        if hasattr(args, "profileaug") and args.profileaug is not None:
            profileaug_class = profileaug_choices.get_class(args.profileaug)
            profileaug = profileaug_class(**args.profileaug_conf)
        else:
            profileaug = None

        # 3. Normalization layer
        if args.normalize is not None:
@@ -472,6 +522,7 @@
            vocab_size=vocab_size,
            frontend=frontend,
            specaug=specaug,
            profileaug=profileaug,
            normalize=normalize,
            label_aggregator=label_aggregator,
            encoder=encoder,
@@ -486,8 +537,8 @@
        # 10. Initialize
        if args.init is not None:
            initialize(model, args.init)
            logging.info(f"Init model parameters with {args.init}.")

        assert check_return_type(model)
        return model

    # ~~~~~~~~~ The methods below are mainly used for inference ~~~~~~~~~
@@ -497,7 +548,7 @@
            config_file: Union[Path, str] = None,
            model_file: Union[Path, str] = None,
            cmvn_file: Union[Path, str] = None,
            device: str = "cpu",
            device: Union[str, torch.device] = "cpu",
    ):
        """Build model from the files.

@@ -510,7 +561,6 @@
            device: Device type, "cpu", "cuda", or "cuda:N".

        """
        assert check_argument_types()
        if config_file is None:
            assert model_file is not None, (
                "The argument 'model_file' must be provided "
@@ -526,9 +576,9 @@
            args["cmvn_file"] = cmvn_file
        args = argparse.Namespace(**args)
        model = cls.build_model(args)
        if not isinstance(model, AbsESPnetModel):
        if not isinstance(model, torch.nn.Module):
            raise RuntimeError(
                f"model must inherit {AbsESPnetModel.__name__}, but got {type(model)}"
                f"model must inherit {torch.nn.Module.__name__}, but got {type(model)}"
            )
        model.to(device)
        model_dict = dict()
@@ -549,15 +599,30 @@
                    model_dict = torch.load(model_name_pth, map_location=device)
                else:
                    model_dict = cls.convert_tf2torch(model, model_file)
                model.load_state_dict(model_dict)
                # model.load_state_dict(model_dict)
            else:
                model_dict = torch.load(model_file, map_location=device)
        model_dict = cls.fileter_model_dict(model_dict, model.state_dict())
        model.load_state_dict(model_dict)
        if model_name_pth is not None and not os.path.exists(model_name_pth):
            torch.save(model_dict, model_name_pth)
            logging.info("model_file is saved to pth: {}".format(model_name_pth))

        return model, args

    @classmethod
    def fileter_model_dict(cls, src_dict: dict, dest_dict: dict):
        from collections import OrderedDict
        new_dict = OrderedDict()
        for key, value in src_dict.items():
            if key in dest_dict:
                new_dict[key] = value
            else:
                logging.info("{} is no longer needed in this model.".format(key))
        for key, value in dest_dict.items():
            if key not in new_dict:
                logging.warning("{} is missed in checkpoint.".format(key))
        return new_dict

    @classmethod
    def convert_tf2torch(
@@ -571,19 +636,24 @@
        var_dict_torch = model.state_dict()
        var_dict_torch_update = dict()
        # speech encoder
        var_dict_torch_update_local = model.encoder.convert_tf2torch(var_dict_tf, var_dict_torch)
        var_dict_torch_update.update(var_dict_torch_update_local)
        if model.encoder is not None:
            var_dict_torch_update_local = model.encoder.convert_tf2torch(var_dict_tf, var_dict_torch)
            var_dict_torch_update.update(var_dict_torch_update_local)
        # speaker encoder
        var_dict_torch_update_local = model.speaker_encoder.convert_tf2torch(var_dict_tf, var_dict_torch)
        var_dict_torch_update.update(var_dict_torch_update_local)
        if model.speaker_encoder is not None:
            var_dict_torch_update_local = model.speaker_encoder.convert_tf2torch(var_dict_tf, var_dict_torch)
            var_dict_torch_update.update(var_dict_torch_update_local)
        # cd scorer
        var_dict_torch_update_local = model.cd_scorer.convert_tf2torch(var_dict_tf, var_dict_torch)
        var_dict_torch_update.update(var_dict_torch_update_local)
        if model.cd_scorer is not None:
            var_dict_torch_update_local = model.cd_scorer.convert_tf2torch(var_dict_tf, var_dict_torch)
            var_dict_torch_update.update(var_dict_torch_update_local)
        # ci scorer
        var_dict_torch_update_local = model.ci_scorer.convert_tf2torch(var_dict_tf, var_dict_torch)
        var_dict_torch_update.update(var_dict_torch_update_local)
        if model.ci_scorer is not None:
            var_dict_torch_update_local = model.ci_scorer.convert_tf2torch(var_dict_tf, var_dict_torch)
            var_dict_torch_update.update(var_dict_torch_update_local)
        # decoder
        var_dict_torch_update_local = model.decoder.convert_tf2torch(var_dict_tf, var_dict_torch)
        var_dict_torch_update.update(var_dict_torch_update_local)
        if model.decoder is not None:
            var_dict_torch_update_local = model.decoder.convert_tf2torch(var_dict_tf, var_dict_torch)
            var_dict_torch_update.update(var_dict_torch_update_local)

        return var_dict_torch_update