update repo

嘉渊

2023-06-14 3d70934e7fed7c0d3179fec340761466205cb3e9

update repo

 funasr/bin/asr_infer.py

@@ -1,66 +1,48 @@
# -*- encoding: utf-8 -*-
#!/usr/bin/env python3
# -*- encoding: utf-8 -*-
# Copyright FunASR (https://github.com/alibaba-damo-academy/FunASR). All Rights Reserved.
#  MIT License  (https://opensource.org/licenses/MIT)

import argparse
import logging
import sys
import time

import codecs
import copy
import logging
import os
import re
import codecs
import tempfile
import requests
from pathlib import Path
from typing import Any
from typing import Dict
from typing import List
from typing import Optional
from typing import Sequence
from typing import Tuple
from typing import Union
from typing import Dict
from typing import Any
from typing import List

import numpy as np
import requests
import torch
from packaging.version import parse as V
from typeguard import check_argument_types
from typeguard import check_return_type
from funasr.fileio.datadir_writer import DatadirWriter

from funasr.models.e2e_asr_contextual_paraformer import NeatContextualParaformer
from funasr.models.e2e_asr_paraformer import BiCifParaformer, ContextualParaformer
from funasr.models.frontend.wav_frontend import WavFrontend, WavFrontendOnline
from funasr.modules.beam_search.beam_search import BeamSearch
# from funasr.modules.beam_search.beam_search import BeamSearchPara as BeamSearch
from funasr.modules.beam_search.beam_search import Hypothesis
from funasr.modules.beam_search.beam_search_sa_asr import Hypothesis as HypothesisSAASR
from funasr.modules.beam_search.beam_search_transducer import BeamSearchTransducer
from funasr.modules.beam_search.beam_search_transducer import Hypothesis as HypothesisTransducer
from funasr.modules.beam_search.beam_search_sa_asr import Hypothesis as HypothesisSAASR
from funasr.modules.scorers.ctc import CTCPrefixScorer
from funasr.modules.scorers.length_bonus import LengthBonus
from funasr.modules.subsampling import TooShortUttError
from funasr.tasks.asr import ASRTask
from funasr.tasks.asr import frontend_choices
from funasr.tasks.lm import LMTask
from funasr.text.build_tokenizer import build_tokenizer
from funasr.text.token_id_converter import TokenIDConverter
from funasr.torch_utils.device_funcs import to_device
from funasr.torch_utils.set_all_random_seed import set_all_random_seed
from funasr.utils import config_argparse
from funasr.utils.cli_utils import get_commandline_args
from funasr.utils.types import str2bool
from funasr.utils.types import str2triple_str
from funasr.utils.types import str_or_none
from funasr.utils import asr_utils, wav_utils, postprocess_utils
from funasr.models.frontend.wav_frontend import WavFrontend, WavFrontendOnline
from funasr.models.e2e_asr_paraformer import BiCifParaformer, ContextualParaformer
from funasr.models.e2e_asr_contextual_paraformer import NeatContextualParaformer
from funasr.export.models.e2e_asr_paraformer import Paraformer as Paraformer_export
from funasr.utils.timestamp_tools import ts_prediction_lfr6_standard
from funasr.bin.tp_infer import Speech2Timestamp
from funasr.bin.vad_infer import Speech2VadSegment
from funasr.bin.punc_infer import Text2Punc
from funasr.utils.vad_utils import slice_padding_fbank
from funasr.tasks.vad import VADTask
from funasr.utils.timestamp_tools import time_stamp_sentence, ts_prediction_lfr6_standard
from funasr.tasks.asr import frontend_choices


class Speech2Text:
    """Speech2Text class
@@ -272,6 +254,7 @@
        assert check_return_type(results)
        return results


class Speech2TextParaformer:
    """Speech2Text class

@@ -466,13 +449,16 @@
        pre_token_length = pre_token_length.round().long()
        if torch.max(pre_token_length) < 1:
            return []
        if not isinstance(self.asr_model, ContextualParaformer) and not isinstance(self.asr_model, NeatContextualParaformer):
        if not isinstance(self.asr_model, ContextualParaformer) and not isinstance(self.asr_model,
                                                                                   NeatContextualParaformer):
            if self.hotword_list:
                logging.warning("Hotword is given but asr model is not a ContextualParaformer.")
            decoder_outs = self.asr_model.cal_decoder_with_predictor(enc, enc_len, pre_acoustic_embeds, pre_token_length)
            decoder_outs = self.asr_model.cal_decoder_with_predictor(enc, enc_len, pre_acoustic_embeds,
                                                                     pre_token_length)
            decoder_out, ys_pad_lens = decoder_outs[0], decoder_outs[1]
        else:
            decoder_outs = self.asr_model.cal_decoder_with_predictor(enc, enc_len, pre_acoustic_embeds, pre_token_length, hw_list=self.hotword_list)
            decoder_outs = self.asr_model.cal_decoder_with_predictor(enc, enc_len, pre_acoustic_embeds,
                                                                     pre_token_length, hw_list=self.hotword_list)
            decoder_out, ys_pad_lens = decoder_outs[0], decoder_outs[1]

        if isinstance(self.asr_model, BiCifParaformer):
@@ -533,7 +519,6 @@
                                                            vad_offset=begin_time)
                results.append((text, token, token_int, hyp, timestamp, enc_len_batch_total, lfr_factor))


        # assert check_return_type(results)
        return results

@@ -590,6 +575,7 @@
        else:
            hotword_list = None
        return hotword_list


class Speech2TextParaformerOnline:
    """Speech2Text class
@@ -845,6 +831,7 @@
        # assert check_return_type(results)
        return results


class Speech2TextUniASR:
    """Speech2Text class