From 2a66366be4c2715870e4859fd5a5db6e8a9dc00a Mon Sep 17 00:00:00 2001
From: chenmengzheAAA <123789350+chenmengzheAAA@users.noreply.github.com>
Date: 星期四, 14 九月 2023 19:00:17 +0800
Subject: [PATCH] Merge pull request #956 from alibaba-damo-academy/chenmengzheAAA-patch-4
---
funasr/datasets/small_datasets/preprocessor.py | 66 ++++++++++++++++++++++++++-------
1 files changed, 52 insertions(+), 14 deletions(-)
diff --git a/funasr/datasets/small_datasets/preprocessor.py b/funasr/datasets/small_datasets/preprocessor.py
index e06a463..0ebf325 100644
--- a/funasr/datasets/small_datasets/preprocessor.py
+++ b/funasr/datasets/small_datasets/preprocessor.py
@@ -10,8 +10,6 @@
import numpy as np
import scipy.signal
import soundfile
-from typeguard import check_argument_types
-from typeguard import check_return_type
from funasr.text.build_tokenizer import build_tokenizer
from funasr.text.cleaner import TextCleaner
@@ -260,7 +258,6 @@
def _speech_process(
self, data: Dict[str, Union[str, np.ndarray]]
) -> Dict[str, Union[str, np.ndarray]]:
- assert check_argument_types()
if self.speech_name in data:
if self.train and (self.rirs is not None or self.noises is not None):
speech = data[self.speech_name]
@@ -347,7 +344,6 @@
speech = data[self.speech_name]
ma = np.max(np.abs(speech))
data[self.speech_name] = speech * self.speech_volume_normalize / ma
- assert check_return_type(data)
return data
def _text_process(
@@ -365,13 +361,11 @@
tokens = self.tokenizer.text2tokens(text)
text_ints = self.token_id_converter.tokens2ids(tokens)
data[self.text_name] = np.array(text_ints, dtype=np.int64)
- assert check_return_type(data)
return data
def __call__(
self, uid: str, data: Dict[str, Union[str, np.ndarray]]
) -> Dict[str, np.ndarray]:
- assert check_argument_types()
data = self._speech_process(data)
data = self._text_process(data)
@@ -439,7 +433,6 @@
tokens = self.tokenizer.text2tokens(text)
text_ints = self.token_id_converter.tokens2ids(tokens)
data[self.text_name] = np.array(text_ints, dtype=np.int64)
- assert check_return_type(data)
return data
@@ -496,13 +489,11 @@
tokens = self.tokenizer.text2tokens(text)
text_ints = self.token_id_converter.tokens2ids(tokens)
data[text_n] = np.array(text_ints, dtype=np.int64)
- assert check_return_type(data)
return data
def __call__(
self, uid: str, data: Dict[str, Union[str, np.ndarray]]
) -> Dict[str, np.ndarray]:
- assert check_argument_types()
if self.speech_name in data:
# Nothing now: candidates:
@@ -606,7 +597,6 @@
tokens = self.tokenizer[i].text2tokens(text)
text_ints = self.token_id_converter[i].tokens2ids(tokens)
data[text_name] = np.array(text_ints, dtype=np.int64)
- assert check_return_type(data)
return data
@@ -685,7 +675,6 @@
def __call__(
self, uid: str, data: Dict[str, Union[list, str, np.ndarray]]
) -> Dict[str, Union[list, np.ndarray]]:
- assert check_argument_types()
# Split words.
if isinstance(data[self.text_name], str):
split_text = self.split_words(data[self.text_name])
@@ -819,8 +808,57 @@
return sentences
-def build_preprocess(args):
- if args.task_name == "asr":
- pass
+def build_preprocess(args, train):
+ if not args.use_preprocessor:
+ return None
+ if args.task_name in ["asr", "data2vec", "diar", "sv"]:
+ retval = CommonPreprocessor(
+ train=train,
+ token_type=args.token_type,
+ token_list=args.token_list,
+ bpemodel=args.bpemodel,
+ non_linguistic_symbols=args.non_linguistic_symbols if hasattr(args, "non_linguistic_symbols") else None,
+ text_cleaner=args.cleaner,
+ g2p_type=args.g2p,
+ split_with_space=args.split_with_space if hasattr(args, "split_with_space") else False,
+ seg_dict_file=args.seg_dict_file if hasattr(args, "seg_dict_file") else None,
+ rir_scp=args.rir_scp if hasattr(args, "rir_scp") else None,
+ rir_apply_prob=args.rir_apply_prob if hasattr(args, "rir_apply_prob") else 1.0,
+ noise_scp=args.noise_scp if hasattr(args, "noise_scp") else None,
+ noise_apply_prob=args.noise_apply_prob if hasattr(args, "noise_apply_prob") else 1.0,
+ noise_db_range=args.noise_db_range if hasattr(args, "noise_db_range") else "13_15",
+ speech_volume_normalize=args.speech_volume_normalize if hasattr(args, "rir_scp") else None,
+ )
+ elif args.task_name == "punc":
+ token_types = [args.token_type, args.token_type]
+ token_lists = [args.token_list, args.punc_list]
+ bpemodels = [args.bpemodel, args.bpemodel]
+ text_names = ["text", "punc"]
+ retval = PuncTrainTokenizerCommonPreprocessor(
+ train=train,
+ token_type=token_types,
+ token_list=token_lists,
+ bpemodel=bpemodels,
+ text_cleaner=args.cleaner,
+ g2p_type=args.g2p,
+ text_name=text_names,
+ non_linguistic_symbols=args.non_linguistic_symbols,
+ )
+ elif args.task_name == "lm":
+ retval = LMPreprocessor(
+ train=train,
+ token_type=args.token_type,
+ token_list=args.token_list,
+ bpemodel=args.bpemodel,
+ text_cleaner=args.cleaner,
+ g2p_type=args.g2p,
+ text_name="text",
+ non_linguistic_symbols=args.non_linguistic_symbols,
+ split_with_space=args.split_with_space,
+ seg_dict_file=args.seg_dict_file
+ )
+ elif args.task_name == "vad":
+ retval = None
else:
raise ValueError(f"Not supported task={args.task_name}")
+ return retval
--
Gitblit v1.9.1