fsmn-vad bugfix (#1270)

zhifu gao

2024-01-18 b28f3c9da94ae72a3a0b7bb5982b587be7cf4cd6

fsmn-vad bugfix (#1270)

* funasr1.0 funetine

* funasr1.0 pbar

* update with main (#1260)

* Update websocket_protocol_zh.md

* update

---------

Co-authored-by: Yabin Li <wucong.lyb@alibaba-inc.com>
Co-authored-by: shixian.shi <shixian.shi@alibaba-inc.com>

* update with main (#1264)

* Funasr1.0 (#1261)

* funasr1.0 funetine

* funasr1.0 pbar

* update with main (#1260)

* Update websocket_protocol_zh.md

* update

---------

Co-authored-by: Yabin Li <wucong.lyb@alibaba-inc.com>
Co-authored-by: shixian.shi <shixian.shi@alibaba-inc.com>

---------

Co-authored-by: Yabin Li <wucong.lyb@alibaba-inc.com>
Co-authored-by: shixian.shi <shixian.shi@alibaba-inc.com>

* bug fix

---------

Co-authored-by: Yabin Li <wucong.lyb@alibaba-inc.com>
Co-authored-by: shixian.shi <shixian.shi@alibaba-inc.com>

* funasr1.0 sanm scama

* funasr1.0 infer_after_finetune

* funasr1.0 fsmn-vad bug fix

* funasr1.0 fsmn-vad bug fix

---------

Co-authored-by: Yabin Li <wucong.lyb@alibaba-inc.com>
Co-authored-by: shixian.shi <shixian.shi@alibaba-inc.com>

 README.md

@@ -91,12 +91,13 @@
from funasr import AutoModel
# paraformer-zh is a multi-functional asr model
# use vad, punc, spk or not as you need
model = AutoModel(model="paraformer-zh", model_revision="v2.0.2", \
                  vad_model="fsmn-vad", vad_model_revision="v2.0.2", \
                  punc_model="ct-punc-c", punc_model_revision="v2.0.2", \
                  spk_model="cam++", spk_model_revision="v2.0.2")
model = AutoModel(model="paraformer-zh", model_revision="v2.0.2",
                  vad_model="fsmn-vad", vad_model_revision="v2.0.2",
                  punc_model="ct-punc-c", punc_model_revision="v2.0.2",
                  # spk_model="cam++", spk_model_revision="v2.0.2",
                  )
res = model.generate(input=f"{model.model_path}/example/asr_example.wav", 
                     batch_size=64, 
                     batch_size_s=300, 
                     hotword='魔搭')
print(res)
```

 README_zh.md

@@ -87,12 +87,13 @@
from funasr import AutoModel
# paraformer-zh is a multi-functional asr model
# use vad, punc, spk or not as you need
model = AutoModel(model="paraformer-zh", model_revision="v2.0.2", \
                  vad_model="fsmn-vad", vad_model_revision="v2.0.2", \
                  punc_model="ct-punc-c", punc_model_revision="v2.0.2", \
                  spk_model="cam++", spk_model_revision="v2.0.2")
model = AutoModel(model="paraformer-zh", model_revision="v2.0.2",
                  vad_model="fsmn-vad", vad_model_revision="v2.0.2",
                  punc_model="ct-punc-c", punc_model_revision="v2.0.2",
                  # spk_model="cam++", spk_model_revision="v2.0.2",
                  )
res = model.generate(input=f"{model.model_path}/example/asr_example.wav", 
            batch_size=64, 
            batch_size_s=300, 
            hotword='魔搭')
print(res)
```

 examples/industrial_data_pretraining/paraformer/infer_after_finetune.sh

New file
@@ -0,0 +1,12 @@


python funasr/bin/inference.py \
--config-path="/Users/zhifu/funasr_github/test_local/funasr_cli_egs" \
--config-name="config.yaml" \
++init_param="/Users/zhifu/funasr_github/test_local/funasr_cli_egs/model.pt" \
+tokenizer_conf.token_list="/Users/zhifu/funasr_github/test_local/funasr_cli_egs/tokens.txt" \
+frontend_conf.cmvn_file="/Users/zhifu/funasr_github/test_local/funasr_cli_egs/am.mvn" \
+input="data/wav.scp" \
+output_dir="./outputs/debug" \
+device="cuda" \


 examples/industrial_data_pretraining/scama/demo.py

New file
@@ -0,0 +1,42 @@
#!/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)

from funasr import AutoModel

chunk_size = [5, 10, 5] #[0, 10, 5] 600ms, [0, 8, 4] 480ms
encoder_chunk_look_back = 0 #number of chunks to lookback for encoder self-attention
decoder_chunk_look_back = 0 #number of encoder chunks to lookback for decoder cross-attention

model = AutoModel(model="/Users/zhifu/Downloads/modelscope_models/speech_SCAMA_asr-zh-cn-16k-common-vocab8358-streaming", model_revision="v2.0.2")
cache = {}
res = model.generate(input="https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav",
            chunk_size=chunk_size,
            encoder_chunk_look_back=encoder_chunk_look_back,
            decoder_chunk_look_back=decoder_chunk_look_back,
            )
print(res)


import soundfile
import os

wav_file = os.path.join(model.model_path, "example/asr_example.wav")
speech, sample_rate = soundfile.read(wav_file)

chunk_stride = chunk_size[1] * 960 # 600ms、480ms

cache = {}
total_chunk_num = int(len((speech)-1)/chunk_stride+1)
for i in range(total_chunk_num):
    speech_chunk = speech[i*chunk_stride:(i+1)*chunk_stride]
    is_final = i == total_chunk_num - 1
    res = model.generate(input=speech_chunk,
                         cache=cache,
                         is_final=is_final,
                         chunk_size=chunk_size,
                         encoder_chunk_look_back=encoder_chunk_look_back,
                         decoder_chunk_look_back=decoder_chunk_look_back,
                         )
    print(res)

 examples/industrial_data_pretraining/scama/infer.sh

New file
@@ -0,0 +1,11 @@

model="damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online"
model_revision="v2.0.2"

python funasr/bin/inference.py \
+model=${model} \
+model_revision=${model_revision} \
+input="https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav" \
+output_dir="./outputs/debug" \
+device="cpu" \


 funasr/models/fsmn_vad_streaming/model.py

@@ -220,38 +220,41 @@
    def FrameSizeMs(self) -> int:
        return int(self.frame_size_ms)

class Stats(object):
    def __init__(self,
                 sil_pdf_ids,
                 max_end_sil_frame_cnt_thresh,
                 speech_noise_thres,
                 ):

@dataclass
class StatsItem:
        self.data_buf_start_frame = 0
        self.frm_cnt = 0
        self.latest_confirmed_speech_frame = 0
        self.lastest_confirmed_silence_frame = -1
        self.continous_silence_frame_count = 0
        self.vad_state_machine = VadStateMachine.kVadInStateStartPointNotDetected
        self.confirmed_start_frame = -1
        self.confirmed_end_frame = -1
        self.number_end_time_detected = 0
        self.sil_frame = 0
        self.sil_pdf_ids = sil_pdf_ids
        self.noise_average_decibel = -100.0
        self.pre_end_silence_detected = False
        self.next_seg = True
    
    # init variables
    data_buf_start_frame = 0
    frm_cnt = 0
    latest_confirmed_speech_frame = 0
    lastest_confirmed_silence_frame = -1
    continous_silence_frame_count = 0
    vad_state_machine = VadStateMachine.kVadInStateStartPointNotDetected
    confirmed_start_frame = -1
    confirmed_end_frame = -1
    number_end_time_detected = 0
    sil_frame = 0
    sil_pdf_ids: list
    noise_average_decibel = -100.0
    pre_end_silence_detected = False
    next_seg = True # unused
        self.output_data_buf = []
        self.output_data_buf_offset = 0
        self.frame_probs = []
        self.max_end_sil_frame_cnt_thresh = max_end_sil_frame_cnt_thresh
        self.speech_noise_thres = speech_noise_thres
        self.scores = None
        self.max_time_out = False
        self.decibel = []
        self.data_buf = None
        self.data_buf_all = None
        self.waveform = None
        self.last_drop_frames = 0
    
    output_data_buf = []
    output_data_buf_offset = 0
    frame_probs = [] # unused
    max_end_sil_frame_cnt_thresh: int
    speech_noise_thres: float
    scores = None
    max_time_out = False #unused
    decibel = []
    data_buf = None
    data_buf_all = None
    waveform = None
    last_drop_frames = 0
    
@tables.register("model_classes", "FsmnVADStreaming")
class FsmnVADStreaming(nn.Module):
@@ -506,10 +509,11 @@
                                          self.vad_opts.sil_to_speech_time_thres,
                                          self.vad_opts.speech_to_sil_time_thres,
                                          self.vad_opts.frame_in_ms)
        windows_detector.Reset()

        stats = StatsItem(sil_pdf_ids=self.vad_opts.sil_pdf_ids,
        stats = Stats(sil_pdf_ids=self.vad_opts.sil_pdf_ids,
                          max_end_sil_frame_cnt_thresh=self.vad_opts.max_end_silence_time - self.vad_opts.speech_to_sil_time_thres,
                          speech_noise_thres=self.vad_opts.speech_noise_thres,
                      speech_noise_thres=self.vad_opts.speech_noise_thres
                      )
        cache["windows_detector"] = windows_detector
        cache["stats"] = stats
@@ -579,7 +583,7 @@

        cache["prev_samples"] = audio_sample[:-m]
        if _is_final:
            self.init_cache(cache, **kwargs)
            cache = {}

        ibest_writer = None
        if ibest_writer is None and kwargs.get("output_dir") is not None:

 funasr/models/paraformer/model.py

@@ -33,7 +33,6 @@
    
    def __init__(
        self,
        # token_list: Union[Tuple[str, ...], List[str]],
        specaug: Optional[str] = None,
        specaug_conf: Optional[Dict] = None,
        normalize: str = None,

 funasr/models/paraformer/template.yaml

@@ -6,7 +6,6 @@
# tables.print()

# network architecture
#model: funasr.models.paraformer.model:Paraformer
model: Paraformer
model_conf:
    ctc_weight: 0.0
@@ -87,13 +86,6 @@
  accum_grad: 1
  grad_clip: 5
  max_epoch: 150
  val_scheduler_criterion:
      - valid
      - acc
  best_model_criterion:
  -   - valid
      - acc
      - max
  keep_nbest_models: 10
  avg_nbest_model: 5
  log_interval: 50

 funasr/models/sanm/decoder.py

@@ -1,3 +1,8 @@
#!/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)

from typing import List
from typing import Tuple
import logging
@@ -193,10 +198,9 @@
@tables.register("decoder_classes", "FsmnDecoder")
class FsmnDecoder(BaseTransformerDecoder):
    """
    Author: Speech Lab of DAMO Academy, Alibaba Group
    SCAMA: Streaming chunk-aware multihead attention for online end-to-end speech recognition
    Author: Zhifu Gao, Shiliang Zhang, Ming Lei, Ian McLoughlin
    San-m: Memory equipped self-attention for end-to-end speech recognition
    https://arxiv.org/abs/2006.01713

    """
    
    def __init__(

 funasr/models/sanm/encoder.py

@@ -1,3 +1,8 @@
#!/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)

from typing import List
from typing import Optional
from typing import Sequence
@@ -156,10 +161,9 @@
@tables.register("encoder_classes", "SANMEncoder")
class SANMEncoder(nn.Module):
    """
    Author: Speech Lab of DAMO Academy, Alibaba Group
    Author: Zhifu Gao, Shiliang Zhang, Ming Lei, Ian McLoughlin
    San-m: Memory equipped self-attention for end-to-end speech recognition
    https://arxiv.org/abs/2006.01713

    """

    def __init__(

 funasr/models/sanm/model.py

@@ -1,3 +1,8 @@
#!/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 logging

import torch
@@ -7,7 +12,11 @@

@tables.register("model_classes", "SANM")
class SANM(Transformer):
    """CTC-attention hybrid Encoder-Decoder model"""
    """
    Author: Zhifu Gao, Shiliang Zhang, Ming Lei, Ian McLoughlin
    San-m: Memory equipped self-attention for end-to-end speech recognition
    https://arxiv.org/abs/2006.01713
    """

    def __init__(
        self,

 funasr/models/sanm/template.yaml

New file
@@ -0,0 +1,121 @@
# This is an example that demonstrates how to configure a model file.
# You can modify the configuration according to your own requirements.

# to print the register_table:
# from funasr.register import tables
# tables.print()

# network architecture
model: SANM
model_conf:
    ctc_weight: 0.0
    lsm_weight: 0.1
    length_normalized_loss: true

# encoder
encoder: SANMEncoder
encoder_conf:
    output_size: 512
    attention_heads: 4
    linear_units: 2048
    num_blocks: 50
    dropout_rate: 0.1
    positional_dropout_rate: 0.1
    attention_dropout_rate: 0.1
    input_layer: pe
    pos_enc_class: SinusoidalPositionEncoder
    normalize_before: true
    kernel_size: 11
    sanm_shfit: 0
    selfattention_layer_type: sanm

# decoder
decoder: FsmnDecoder
decoder_conf:
    attention_heads: 4
    linear_units: 2048
    num_blocks: 16
    dropout_rate: 0.1
    positional_dropout_rate: 0.1
    self_attention_dropout_rate: 0.1
    src_attention_dropout_rate: 0.1
    att_layer_num: 16
    kernel_size: 11
    sanm_shfit: 0



# frontend related
frontend: WavFrontend
frontend_conf:
    fs: 16000
    window: hamming
    n_mels: 80
    frame_length: 25
    frame_shift: 10
    lfr_m: 7
    lfr_n: 6

specaug: SpecAugLFR
specaug_conf:
    apply_time_warp: false
    time_warp_window: 5
    time_warp_mode: bicubic
    apply_freq_mask: true
    freq_mask_width_range:
    - 0
    - 30
    lfr_rate: 6
    num_freq_mask: 1
    apply_time_mask: true
    time_mask_width_range:
    - 0
    - 12
    num_time_mask: 1

train_conf:
  accum_grad: 1
  grad_clip: 5
  max_epoch: 150
  val_scheduler_criterion:
      - valid
      - acc
  best_model_criterion:
  -   - valid
      - acc
      - max
  keep_nbest_models: 10
  avg_nbest_model: 5
  log_interval: 50

optim: adam
optim_conf:
   lr: 0.0005
scheduler: warmuplr
scheduler_conf:
   warmup_steps: 30000

dataset: AudioDataset
dataset_conf:
    index_ds: IndexDSJsonl
    batch_sampler: DynamicBatchLocalShuffleSampler
    batch_type: example # example or length
    batch_size: 1 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
    max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
    buffer_size: 500
    shuffle: True
    num_workers: 0

tokenizer: CharTokenizer
tokenizer_conf:
  unk_symbol: <unk>
  split_with_space: true


ctc_conf:
    dropout_rate: 0.0
    ctc_type: builtin
    reduce: true
    ignore_nan_grad: true

normalize: null

 funasr/models/scama/decoder.py

File was renamed from funasr/models/scama/sanm_decoder.py
@@ -1,3 +1,8 @@
#!/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)

from typing import List
from typing import Tuple
import logging
@@ -192,11 +197,11 @@
@tables.register("decoder_classes", "FsmnDecoderSCAMAOpt")
class FsmnDecoderSCAMAOpt(BaseTransformerDecoder):
    """
    Author: Speech Lab of DAMO Academy, Alibaba Group
    Author: Shiliang Zhang, Zhifu Gao, Haoneng Luo, Ming Lei, Jie Gao, Zhijie Yan, Lei Xie
    SCAMA: Streaming chunk-aware multihead attention for online end-to-end speech recognition
    https://arxiv.org/abs/2006.01713

    https://arxiv.org/abs/2006.01712
    """
    
    def __init__(
            self,
            vocab_size: int,

 funasr/models/scama/encoder.py

File was renamed from funasr/models/scama/sanm_encoder.py
@@ -1,3 +1,8 @@
#!/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)

from typing import List
from typing import Optional
from typing import Sequence
@@ -157,10 +162,9 @@
@tables.register("encoder_classes", "SANMEncoderChunkOpt")
class SANMEncoderChunkOpt(nn.Module):
    """
    Author: Speech Lab of DAMO Academy, Alibaba Group
    Author: Shiliang Zhang, Zhifu Gao, Haoneng Luo, Ming Lei, Jie Gao, Zhijie Yan, Lei Xie
    SCAMA: Streaming chunk-aware multihead attention for online end-to-end speech recognition
    https://arxiv.org/abs/2006.01713

    https://arxiv.org/abs/2006.01712
    """

    def __init__(

 funasr/models/scama/model.py

New file
@@ -0,0 +1,669 @@
#!/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 time
import torch
import torch.nn as nn
import torch.functional as F
import logging
from typing import Dict, Tuple
from contextlib import contextmanager
from distutils.version import LooseVersion

from funasr.register import tables
from funasr.models.ctc.ctc import CTC
from funasr.utils import postprocess_utils
from funasr.metrics.compute_acc import th_accuracy
from funasr.utils.datadir_writer import DatadirWriter
from funasr.models.paraformer.model import Paraformer
from funasr.models.paraformer.search import Hypothesis
from funasr.models.paraformer.cif_predictor import mae_loss
from funasr.train_utils.device_funcs import force_gatherable
from funasr.losses.label_smoothing_loss import LabelSmoothingLoss
from funasr.models.transformer.utils.add_sos_eos import add_sos_eos
from funasr.models.transformer.utils.nets_utils import make_pad_mask, pad_list
from funasr.utils.load_utils import load_audio_text_image_video, extract_fbank
from funasr.models.scama.utils import sequence_mask

if LooseVersion(torch.__version__) >= LooseVersion("1.6.0"):
    from torch.cuda.amp import autocast
else:
    # Nothing to do if torch<1.6.0
    @contextmanager
    def autocast(enabled=True):
        yield

@tables.register("model_classes", "SCAMA")
class SCAMA(nn.Module):
    """
    Author: Shiliang Zhang, Zhifu Gao, Haoneng Luo, Ming Lei, Jie Gao, Zhijie Yan, Lei Xie
    SCAMA: Streaming chunk-aware multihead attention for online end-to-end speech recognition
    https://arxiv.org/abs/2006.01712
    """

    def __init__(
        self,
        specaug: str = None,
        specaug_conf: dict = None,
        normalize: str = None,
        normalize_conf: dict = None,
        encoder: str = None,
        encoder_conf: dict = None,
        decoder: str = None,
        decoder_conf: dict = None,
        ctc: str = None,
        ctc_conf: dict = None,
        ctc_weight: float = 0.5,
        predictor: str = None,
        predictor_conf: dict = None,
        predictor_bias: int = 0,
        predictor_weight: float = 0.0,
        input_size: int = 80,
        vocab_size: int = -1,
        ignore_id: int = -1,
        blank_id: int = 0,
        sos: int = 1,
        eos: int = 2,
        lsm_weight: float = 0.0,
        length_normalized_loss: bool = False,
        share_embedding: bool = False,
        **kwargs,
    ):

        super().__init__()

        if specaug is not None:
            specaug_class = tables.specaug_classes.get(specaug)
            specaug = specaug_class(**specaug_conf)
            
        if normalize is not None:
            normalize_class = tables.normalize_classes.get(normalize)
            normalize = normalize_class(**normalize_conf)
            
        encoder_class = tables.encoder_classes.get(encoder)
        encoder = encoder_class(input_size=input_size, **encoder_conf)
        encoder_output_size = encoder.output_size()

        decoder_class = tables.decoder_classes.get(decoder)
        decoder = decoder_class(
            vocab_size=vocab_size,
            encoder_output_size=encoder_output_size,
            **decoder_conf,
        )
        if ctc_weight > 0.0:
    
            if ctc_conf is None:
                ctc_conf = {}
    
            ctc = CTC(
                odim=vocab_size, encoder_output_size=encoder_output_size, **ctc_conf
            )

        predictor_class = tables.predictor_classes.get(predictor)
        predictor = predictor_class(**predictor_conf)

        # note that eos is the same as sos (equivalent ID)
        self.blank_id = blank_id
        self.sos = sos if sos is not None else vocab_size - 1
        self.eos = eos if eos is not None else vocab_size - 1
        self.vocab_size = vocab_size
        self.ignore_id = ignore_id
        self.ctc_weight = ctc_weight
        
        self.specaug = specaug
        self.normalize = normalize
        
        self.encoder = encoder


        if ctc_weight == 1.0:
            self.decoder = None
        else:
            self.decoder = decoder

        self.criterion_att = LabelSmoothingLoss(
            size=vocab_size,
            padding_idx=ignore_id,
            smoothing=lsm_weight,
            normalize_length=length_normalized_loss,
        )

        if ctc_weight == 0.0:
            self.ctc = None
        else:
            self.ctc = ctc
            
        self.predictor = predictor
        self.predictor_weight = predictor_weight
        self.predictor_bias = predictor_bias

        self.criterion_pre = mae_loss(normalize_length=length_normalized_loss)

        self.share_embedding = share_embedding
        if self.share_embedding:
            self.decoder.embed = None

        self.length_normalized_loss = length_normalized_loss
        self.beam_search = None
        self.error_calculator = None
        
        if self.encoder.overlap_chunk_cls is not None:
            from funasr.models.scama.chunk_utilis import build_scama_mask_for_cross_attention_decoder
            self.build_scama_mask_for_cross_attention_decoder_fn = build_scama_mask_for_cross_attention_decoder
            self.decoder_attention_chunk_type = kwargs.get("decoder_attention_chunk_type", "chunk")

    def forward(
        self,
        speech: torch.Tensor,
        speech_lengths: torch.Tensor,
        text: torch.Tensor,
        text_lengths: torch.Tensor,
        **kwargs,
    ) -> Tuple[torch.Tensor, Dict[str, torch.Tensor], torch.Tensor]:
        """Encoder + Decoder + Calc loss
        Args:
                speech: (Batch, Length, ...)
                speech_lengths: (Batch, )
                text: (Batch, Length)
                text_lengths: (Batch,)
        """

        decoding_ind = kwargs.get("decoding_ind")
        if len(text_lengths.size()) > 1:
            text_lengths = text_lengths[:, 0]
        if len(speech_lengths.size()) > 1:
            speech_lengths = speech_lengths[:, 0]
    
        batch_size = speech.shape[0]
    
        # Encoder
        ind = self.encoder.overlap_chunk_cls.random_choice(self.training, decoding_ind)
        encoder_out, encoder_out_lens = self.encode(speech, speech_lengths, ind=ind)

    
        loss_ctc, cer_ctc = None, None
        loss_pre = None
        stats = dict()
    
        # decoder: CTC branch
    
        if self.ctc_weight > 0.0:

            encoder_out_ctc, encoder_out_lens_ctc = self.encoder.overlap_chunk_cls.remove_chunk(encoder_out,
                                                                                                encoder_out_lens,
                                                                                                chunk_outs=None)

        
            loss_ctc, cer_ctc = self._calc_ctc_loss(
                encoder_out_ctc, encoder_out_lens_ctc, text, text_lengths
            )
            # Collect CTC branch stats
            stats["loss_ctc"] = loss_ctc.detach() if loss_ctc is not None else None
            stats["cer_ctc"] = cer_ctc
    
        # decoder: Attention decoder branch
        loss_att, acc_att, cer_att, wer_att, loss_pre = self._calc_att_predictor_loss(
            encoder_out, encoder_out_lens, text, text_lengths
        )
    
        # 3. CTC-Att loss definition
        if self.ctc_weight == 0.0:
            loss = loss_att + loss_pre * self.predictor_weight
        else:
            loss = self.ctc_weight * loss_ctc + (
                1 - self.ctc_weight) * loss_att + loss_pre * self.predictor_weight
    
        # Collect Attn branch stats
        stats["loss_att"] = loss_att.detach() if loss_att is not None else None
        stats["acc"] = acc_att
        stats["cer"] = cer_att
        stats["wer"] = wer_att
        stats["loss_pre"] = loss_pre.detach().cpu() if loss_pre is not None else None
    
        stats["loss"] = torch.clone(loss.detach())
    
        # force_gatherable: to-device and to-tensor if scalar for DataParallel
        if self.length_normalized_loss:
            batch_size = (text_lengths + self.predictor_bias).sum()
        loss, stats, weight = force_gatherable((loss, stats, batch_size), loss.device)
        return loss, stats, weight

    def encode(
        self, speech: torch.Tensor, speech_lengths: torch.Tensor, **kwargs,
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        """Encoder. Note that this method is used by asr_inference.py
        Args:
                speech: (Batch, Length, ...)
                speech_lengths: (Batch, )
                ind: int
        """
        with autocast(False):
        
            # Data augmentation
            if self.specaug is not None and self.training:
                speech, speech_lengths = self.specaug(speech, speech_lengths)
        
            # Normalization for feature: e.g. Global-CMVN, Utterance-CMVN
            if self.normalize is not None:
                speech, speech_lengths = self.normalize(speech, speech_lengths)
    
        # Forward encoder
        encoder_out, encoder_out_lens, _ = self.encoder(speech, speech_lengths)
        if isinstance(encoder_out, tuple):
            encoder_out = encoder_out[0]
    
        return encoder_out, encoder_out_lens

    def encode_chunk(
        self, speech: torch.Tensor, speech_lengths: torch.Tensor, cache: dict = None, **kwargs,
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        """Frontend + Encoder. Note that this method is used by asr_inference.py
        Args:
                speech: (Batch, Length, ...)
                speech_lengths: (Batch, )
                ind: int
        """
        with autocast(False):
        
            # Data augmentation
            if self.specaug is not None and self.training:
                speech, speech_lengths = self.specaug(speech, speech_lengths)
        
            # Normalization for feature: e.g. Global-CMVN, Utterance-CMVN
            if self.normalize is not None:
                speech, speech_lengths = self.normalize(speech, speech_lengths)
    
        # Forward encoder
        encoder_out, encoder_out_lens, _ = self.encoder.forward_chunk(speech, speech_lengths, cache=cache["encoder"])
        if isinstance(encoder_out, tuple):
            encoder_out = encoder_out[0]
    
        return encoder_out, torch.tensor([encoder_out.size(1)])

    def calc_predictor_chunk(self, encoder_out, encoder_out_lens, cache=None, **kwargs):
        is_final = kwargs.get("is_final", False)

        return self.predictor.forward_chunk(encoder_out, cache["encoder"], is_final=is_final)

    def _calc_att_predictor_loss(
        self,
        encoder_out: torch.Tensor,
        encoder_out_lens: torch.Tensor,
        ys_pad: torch.Tensor,
        ys_pad_lens: torch.Tensor,
    ):
        ys_in_pad, ys_out_pad = add_sos_eos(ys_pad, self.sos, self.eos, self.ignore_id)
        ys_in_lens = ys_pad_lens + 1

        encoder_out_mask = sequence_mask(encoder_out_lens, maxlen=encoder_out.size(1), dtype=encoder_out.dtype,
                                         device=encoder_out.device)[:, None, :]
        mask_chunk_predictor = None
        if self.encoder.overlap_chunk_cls is not None:
            mask_chunk_predictor = self.encoder.overlap_chunk_cls.get_mask_chunk_predictor(None,
                                                                                           device=encoder_out.device,
                                                                                           batch_size=encoder_out.size(
                                                                                               0))
            mask_shfit_chunk = self.encoder.overlap_chunk_cls.get_mask_shfit_chunk(None, device=encoder_out.device,
                                                                                   batch_size=encoder_out.size(0))
            encoder_out = encoder_out * mask_shfit_chunk
        pre_acoustic_embeds, pre_token_length, pre_alphas, _ = self.predictor(encoder_out,
                                                                              ys_out_pad,
                                                                              encoder_out_mask,
                                                                              ignore_id=self.ignore_id,
                                                                              mask_chunk_predictor=mask_chunk_predictor,
                                                                              target_label_length=ys_in_lens,
                                                                              )
        predictor_alignments, predictor_alignments_len = self.predictor.gen_frame_alignments(pre_alphas,
                                                                                             encoder_out_lens)


        encoder_chunk_size = self.encoder.overlap_chunk_cls.chunk_size_pad_shift_cur
        attention_chunk_center_bias = 0
        attention_chunk_size = encoder_chunk_size
        decoder_att_look_back_factor = self.encoder.overlap_chunk_cls.decoder_att_look_back_factor_cur
        mask_shift_att_chunk_decoder = self.encoder.overlap_chunk_cls.get_mask_shift_att_chunk_decoder(None,
                                                                                                       device=encoder_out.device,
                                                                                                       batch_size=encoder_out.size(
                                                                                                           0))
        scama_mask = self.build_scama_mask_for_cross_attention_decoder_fn(
            predictor_alignments=predictor_alignments,
            encoder_sequence_length=encoder_out_lens,
            chunk_size=1,
            encoder_chunk_size=encoder_chunk_size,
            attention_chunk_center_bias=attention_chunk_center_bias,
            attention_chunk_size=attention_chunk_size,
            attention_chunk_type=self.decoder_attention_chunk_type,
            step=None,
            predictor_mask_chunk_hopping=mask_chunk_predictor,
            decoder_att_look_back_factor=decoder_att_look_back_factor,
            mask_shift_att_chunk_decoder=mask_shift_att_chunk_decoder,
            target_length=ys_in_lens,
            is_training=self.training,
        )


        # try:
        # 1. Forward decoder
        decoder_out, _ = self.decoder(
            encoder_out,
            encoder_out_lens,
            ys_in_pad,
            ys_in_lens,
            chunk_mask=scama_mask,
            pre_acoustic_embeds=pre_acoustic_embeds,

        )

        # 2. Compute attention loss
        loss_att = self.criterion_att(decoder_out, ys_out_pad)
        acc_att = th_accuracy(
            decoder_out.view(-1, self.vocab_size),
            ys_out_pad,
            ignore_label=self.ignore_id,
        )
        # predictor loss
        loss_pre = self.criterion_pre(ys_in_lens.type_as(pre_token_length), pre_token_length)
        # Compute cer/wer using attention-decoder
        if self.training or self.error_calculator is None:
            cer_att, wer_att = None, None
        else:
            ys_hat = decoder_out.argmax(dim=-1)
            cer_att, wer_att = self.error_calculator(ys_hat.cpu(), ys_pad.cpu())

        return loss_att, acc_att, cer_att, wer_att, loss_pre

    def calc_predictor_mask(
        self,
        encoder_out: torch.Tensor,
        encoder_out_lens: torch.Tensor,
        ys_pad: torch.Tensor = None,
        ys_pad_lens: torch.Tensor = None,
    ):
        # ys_in_pad, ys_out_pad = add_sos_eos(ys_pad, self.sos, self.eos, self.ignore_id)
        # ys_in_lens = ys_pad_lens + 1
        ys_out_pad, ys_in_lens = None, None

        encoder_out_mask = sequence_mask(encoder_out_lens, maxlen=encoder_out.size(1), dtype=encoder_out.dtype,
                                         device=encoder_out.device)[:, None, :]
        mask_chunk_predictor = None

        mask_chunk_predictor = self.encoder.overlap_chunk_cls.get_mask_chunk_predictor(None,
                                                                                       device=encoder_out.device,
                                                                                       batch_size=encoder_out.size(
                                                                                           0))
        mask_shfit_chunk = self.encoder.overlap_chunk_cls.get_mask_shfit_chunk(None, device=encoder_out.device,
                                                                               batch_size=encoder_out.size(0))
        encoder_out = encoder_out * mask_shfit_chunk
        pre_acoustic_embeds, pre_token_length, pre_alphas, _ = self.predictor(encoder_out,
                                                                              ys_out_pad,
                                                                              encoder_out_mask,
                                                                              ignore_id=self.ignore_id,
                                                                              mask_chunk_predictor=mask_chunk_predictor,
                                                                              target_label_length=ys_in_lens,
                                                                              )
        predictor_alignments, predictor_alignments_len = self.predictor.gen_frame_alignments(pre_alphas,
                                                                                             encoder_out_lens)
    

        encoder_chunk_size = self.encoder.overlap_chunk_cls.chunk_size_pad_shift_cur
        attention_chunk_center_bias = 0
        attention_chunk_size = encoder_chunk_size
        decoder_att_look_back_factor = self.encoder.overlap_chunk_cls.decoder_att_look_back_factor_cur
        mask_shift_att_chunk_decoder = self.encoder.overlap_chunk_cls.get_mask_shift_att_chunk_decoder(None,
                                                                                                       device=encoder_out.device,
                                                                                                       batch_size=encoder_out.size(
                                                                                                           0))
        scama_mask = self.build_scama_mask_for_cross_attention_decoder_fn(
            predictor_alignments=predictor_alignments,
            encoder_sequence_length=encoder_out_lens,
            chunk_size=1,
            encoder_chunk_size=encoder_chunk_size,
            attention_chunk_center_bias=attention_chunk_center_bias,
            attention_chunk_size=attention_chunk_size,
            attention_chunk_type=self.decoder_attention_chunk_type,
            step=None,
            predictor_mask_chunk_hopping=mask_chunk_predictor,
            decoder_att_look_back_factor=decoder_att_look_back_factor,
            mask_shift_att_chunk_decoder=mask_shift_att_chunk_decoder,
            target_length=ys_in_lens,
            is_training=self.training,
        )

        return pre_acoustic_embeds, pre_token_length, predictor_alignments, predictor_alignments_len, scama_mask

    def init_beam_search(self,
                         **kwargs,
                         ):
        from funasr.models.scama.beam_search import BeamSearchScama
        from funasr.models.transformer.scorers.ctc import CTCPrefixScorer
        from funasr.models.transformer.scorers.length_bonus import LengthBonus
    
        # 1. Build ASR model
        scorers = {}
    
        if self.ctc != None:
            ctc = CTCPrefixScorer(ctc=self.ctc, eos=self.eos)
            scorers.update(
                ctc=ctc
            )
        token_list = kwargs.get("token_list")
        scorers.update(
            decoder=self.decoder,
            length_bonus=LengthBonus(len(token_list)),
        )
    
        # 3. Build ngram model
        # ngram is not supported now
        ngram = None
        scorers["ngram"] = ngram
    
        weights = dict(
            decoder=1.0 - kwargs.get("decoding_ctc_weight"),
            ctc=kwargs.get("decoding_ctc_weight", 0.0),
            lm=kwargs.get("lm_weight", 0.0),
            ngram=kwargs.get("ngram_weight", 0.0),
            length_bonus=kwargs.get("penalty", 0.0),
        )
        beam_search = BeamSearchScama(
            beam_size=kwargs.get("beam_size", 2),
            weights=weights,
            scorers=scorers,
            sos=self.sos,
            eos=self.eos,
            vocab_size=len(token_list),
            token_list=token_list,
            pre_beam_score_key=None if self.ctc_weight == 1.0 else "full",
        )
        # beam_search.to(device=kwargs.get("device", "cpu"), dtype=getattr(torch, kwargs.get("dtype", "float32"))).eval()
        # for scorer in scorers.values():
        #     if isinstance(scorer, torch.nn.Module):
        #         scorer.to(device=kwargs.get("device", "cpu"), dtype=getattr(torch, kwargs.get("dtype", "float32"))).eval()
        self.beam_search = beam_search

    def generate_chunk(self,
                       speech,
                       speech_lengths=None,
                       key: list = None,
                       tokenizer=None,
                       frontend=None,
                       **kwargs,
                       ):
        cache = kwargs.get("cache", {})
        speech = speech.to(device=kwargs["device"])
        speech_lengths = speech_lengths.to(device=kwargs["device"])
    
        # Encoder
        encoder_out, encoder_out_lens = self.encode_chunk(speech, speech_lengths, cache=cache,
                                                          is_final=kwargs.get("is_final", False))
        if isinstance(encoder_out, tuple):
            encoder_out = encoder_out[0]

        # predictor
        predictor_outs = self.calc_predictor_chunk(encoder_out,
                                                   encoder_out_lens,
                                                   cache=cache,
                                                   is_final=kwargs.get("is_final", False),
                                                   )
        pre_acoustic_embeds, pre_token_length, alphas, pre_peak_index = predictor_outs[0], predictor_outs[1], \
                                                                        predictor_outs[2], predictor_outs[3]
        pre_token_length = pre_token_length.round().long()


        if torch.max(pre_token_length) < 1:
            return []
        decoder_outs = self.cal_decoder_with_predictor_chunk(encoder_out,
                                                             encoder_out_lens,
                                                             pre_acoustic_embeds,
                                                             pre_token_length,
                                                             cache=cache
                                                             )
        decoder_out, ys_pad_lens = decoder_outs[0], decoder_outs[1]
    
        results = []
        b, n, d = decoder_out.size()
        if isinstance(key[0], (list, tuple)):
            key = key[0]
        for i in range(b):
            x = encoder_out[i, :encoder_out_lens[i], :]
            am_scores = decoder_out[i, :pre_token_length[i], :]
            if self.beam_search is not None:
                nbest_hyps = self.beam_search(
                    x=x, am_scores=am_scores, maxlenratio=kwargs.get("maxlenratio", 0.0),
                    minlenratio=kwargs.get("minlenratio", 0.0)
                )
            
                nbest_hyps = nbest_hyps[: self.nbest]
            else:
            
                yseq = am_scores.argmax(dim=-1)
                score = am_scores.max(dim=-1)[0]
                score = torch.sum(score, dim=-1)
                # pad with mask tokens to ensure compatibility with sos/eos tokens
                yseq = torch.tensor(
                    [self.sos] + yseq.tolist() + [self.eos], device=yseq.device
                )
                nbest_hyps = [Hypothesis(yseq=yseq, score=score)]
            for nbest_idx, hyp in enumerate(nbest_hyps):
            
                # remove sos/eos and get results
                last_pos = -1
                if isinstance(hyp.yseq, list):
                    token_int = hyp.yseq[1:last_pos]
                else:
                    token_int = hyp.yseq[1:last_pos].tolist()
            
                # remove blank symbol id, which is assumed to be 0
                token_int = list(filter(lambda x: x != self.eos and x != self.sos and x != self.blank_id, token_int))
            
                # Change integer-ids to tokens
                token = tokenizer.ids2tokens(token_int)
                # text = tokenizer.tokens2text(token)
            
                result_i = token
            
                results.extend(result_i)
    
        return results

    def init_cache(self, cache: dict = {}, **kwargs):
        chunk_size = kwargs.get("chunk_size", [0, 10, 5])
        encoder_chunk_look_back = kwargs.get("encoder_chunk_look_back", 0)
        decoder_chunk_look_back = kwargs.get("decoder_chunk_look_back", 0)
        batch_size = 1
    
        enc_output_size = kwargs["encoder_conf"]["output_size"]
        feats_dims = kwargs["frontend_conf"]["n_mels"] * kwargs["frontend_conf"]["lfr_m"]
        cache_encoder = {"start_idx": 0, "cif_hidden": torch.zeros((batch_size, 1, enc_output_size)),
                         "cif_alphas": torch.zeros((batch_size, 1)), "chunk_size": chunk_size,
                         "encoder_chunk_look_back": encoder_chunk_look_back, "last_chunk": False, "opt": None,
                         "feats": torch.zeros((batch_size, chunk_size[0] + chunk_size[2], feats_dims)),
                         "tail_chunk": False}
        cache["encoder"] = cache_encoder
    
        cache_decoder = {"decode_fsmn": None, "decoder_chunk_look_back": decoder_chunk_look_back, "opt": None,
                         "chunk_size": chunk_size}
        cache["decoder"] = cache_decoder
        cache["frontend"] = {}
        cache["prev_samples"] = torch.empty(0)
    
        return cache

    def inference(self,
                  data_in,
                  data_lengths=None,
                  key: list = None,
                  tokenizer=None,
                  frontend=None,
                  cache: dict = {},
                  **kwargs,
                  ):
    
        # init beamsearch
        is_use_ctc = kwargs.get("decoding_ctc_weight", 0.0) > 0.00001 and self.ctc != None
        is_use_lm = kwargs.get("lm_weight", 0.0) > 0.00001 and kwargs.get("lm_file", None) is not None
        if self.beam_search is None and (is_use_lm or is_use_ctc):
            logging.info("enable beam_search")
            self.init_beam_search(**kwargs)
            self.nbest = kwargs.get("nbest", 1)
    
        if len(cache) == 0:
            self.init_cache(cache, **kwargs)
    
        meta_data = {}
        chunk_size = kwargs.get("chunk_size", [0, 10, 5])
        chunk_stride_samples = int(chunk_size[1] * 960)  # 600ms
    
        time1 = time.perf_counter()
        cfg = {"is_final": kwargs.get("is_final", False)}
        audio_sample_list = load_audio_text_image_video(data_in,
                                                        fs=frontend.fs,
                                                        audio_fs=kwargs.get("fs", 16000),
                                                        data_type=kwargs.get("data_type", "sound"),
                                                        tokenizer=tokenizer,
                                                        cache=cfg,
                                                        )
        _is_final = cfg["is_final"]  # if data_in is a file or url, set is_final=True
    
        time2 = time.perf_counter()
        meta_data["load_data"] = f"{time2 - time1:0.3f}"
        assert len(audio_sample_list) == 1, "batch_size must be set 1"
    
        audio_sample = torch.cat((cache["prev_samples"], audio_sample_list[0]))
    
        n = int(len(audio_sample) // chunk_stride_samples + int(_is_final))
        m = int(len(audio_sample) % chunk_stride_samples * (1 - int(_is_final)))
        tokens = []
        for i in range(n):
            kwargs["is_final"] = _is_final and i == n - 1
            audio_sample_i = audio_sample[i * chunk_stride_samples:(i + 1) * chunk_stride_samples]
        
            # extract fbank feats
            speech, speech_lengths = extract_fbank([audio_sample_i], data_type=kwargs.get("data_type", "sound"),
                                                   frontend=frontend, cache=cache["frontend"],
                                                   is_final=kwargs["is_final"])
            time3 = time.perf_counter()
            meta_data["extract_feat"] = f"{time3 - time2:0.3f}"
            meta_data["batch_data_time"] = speech_lengths.sum().item() * frontend.frame_shift * frontend.lfr_n / 1000
        
            tokens_i = self.generate_chunk(speech, speech_lengths, key=key, tokenizer=tokenizer, cache=cache,
                                           frontend=frontend, **kwargs)
            tokens.extend(tokens_i)
    
        text_postprocessed, _ = postprocess_utils.sentence_postprocess(tokens)
    
        result_i = {"key": key[0], "text": text_postprocessed}
        result = [result_i]
    
        cache["prev_samples"] = audio_sample[:-m]
        if _is_final:
            self.init_cache(cache, **kwargs)
    
        if kwargs.get("output_dir"):
            writer = DatadirWriter(kwargs.get("output_dir"))
            ibest_writer = writer[f"{1}best_recog"]
            ibest_writer["token"][key[0]] = " ".join(tokens)
            ibest_writer["text"][key[0]] = text_postprocessed
    
        return result, meta_data

 funasr/models/scama/template.yaml

New file
@@ -0,0 +1,127 @@
# This is an example that demonstrates how to configure a model file.
# You can modify the configuration according to your own requirements.

# to print the register_table:
# from funasr.register import tables
# tables.print()

# network architecture
model: SCAMA
model_conf:
    ctc_weight: 0.0
    lsm_weight: 0.1
    length_normalized_loss: true

# encoder
encoder: SANMEncoderChunkOpt
encoder_conf:
    output_size: 512
    attention_heads: 4
    linear_units: 2048
    num_blocks: 50
    dropout_rate: 0.1
    positional_dropout_rate: 0.1
    attention_dropout_rate: 0.1
    input_layer: pe
    pos_enc_class: SinusoidalPositionEncoder
    normalize_before: true
    kernel_size: 11
    sanm_shfit: 0
    selfattention_layer_type: sanm

# decoder
decoder: FsmnDecoderSCAMAOpt
decoder_conf:
    attention_heads: 4
    linear_units: 2048
    num_blocks: 16
    dropout_rate: 0.1
    positional_dropout_rate: 0.1
    self_attention_dropout_rate: 0.1
    src_attention_dropout_rate: 0.1
    att_layer_num: 16
    kernel_size: 11
    sanm_shfit: 0

predictor: CifPredictorV2
predictor_conf:
    idim: 512
    threshold: 1.0
    l_order: 1
    r_order: 1
    tail_threshold: 0.45

# frontend related
frontend: WavFrontend
frontend_conf:
    fs: 16000
    window: hamming
    n_mels: 80
    frame_length: 25
    frame_shift: 10
    lfr_m: 7
    lfr_n: 6

specaug: SpecAugLFR
specaug_conf:
    apply_time_warp: false
    time_warp_window: 5
    time_warp_mode: bicubic
    apply_freq_mask: true
    freq_mask_width_range:
    - 0
    - 30
    lfr_rate: 6
    num_freq_mask: 1
    apply_time_mask: true
    time_mask_width_range:
    - 0
    - 12
    num_time_mask: 1

train_conf:
  accum_grad: 1
  grad_clip: 5
  max_epoch: 150
  val_scheduler_criterion:
      - valid
      - acc
  best_model_criterion:
  -   - valid
      - acc
      - max
  keep_nbest_models: 10
  avg_nbest_model: 5
  log_interval: 50

optim: adam
optim_conf:
   lr: 0.0005
scheduler: warmuplr
scheduler_conf:
   warmup_steps: 30000

dataset: AudioDataset
dataset_conf:
    index_ds: IndexDSJsonl
    batch_sampler: DynamicBatchLocalShuffleSampler
    batch_type: example # example or length
    batch_size: 1 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
    max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
    buffer_size: 500
    shuffle: True
    num_workers: 0

tokenizer: CharTokenizer
tokenizer_conf:
  unk_symbol: <unk>
  split_with_space: true


ctc_conf:
    dropout_rate: 0.0
    ctc_type: builtin
    reduce: true
    ignore_nan_grad: true

normalize: null

 funasr/models/uniasr/model.py

File was renamed from funasr/models/uniasr/e2e_uni_asr.py
@@ -1,85 +1,73 @@
import logging
from contextlib import contextmanager
from distutils.version import LooseVersion
from typing import Dict
from typing import List
from typing import Optional
from typing import Tuple
from typing import Union
#!/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 time
import torch

from funasr.models.e2e_asr_common import ErrorCalculator
from funasr.metrics.compute_acc import th_accuracy
from funasr.models.transformer.utils.add_sos_eos import add_sos_eos
from funasr.losses.label_smoothing_loss import (
    LabelSmoothingLoss,  # noqa: H301
)
from funasr.models.ctc import CTC
from funasr.models.decoder.abs_decoder import AbsDecoder
from funasr.models.encoder.abs_encoder import AbsEncoder
from funasr.frontends.abs_frontend import AbsFrontend
from funasr.models.postencoder.abs_postencoder import AbsPostEncoder
from funasr.models.preencoder.abs_preencoder import AbsPreEncoder
from funasr.models.specaug.abs_specaug import AbsSpecAug
from funasr.layers.abs_normalize import AbsNormalize
from funasr.train_utils.device_funcs import force_gatherable
from funasr.models.base_model import FunASRModel
from funasr.models.scama.chunk_utilis import sequence_mask
from funasr.models.paraformer.cif_predictor import mae_loss

if LooseVersion(torch.__version__) >= LooseVersion("1.6.0"):
import logging
    from torch.cuda.amp import autocast
else:
    # Nothing to do if torch<1.6.0
    @contextmanager
    def autocast(enabled=True):
        yield
from typing import Union, Dict, List, Tuple, Optional

from funasr.register import tables
from funasr.models.ctc.ctc import CTC
from funasr.utils import postprocess_utils
from funasr.metrics.compute_acc import th_accuracy
from funasr.utils.datadir_writer import DatadirWriter
from funasr.models.paraformer.search import Hypothesis
from funasr.models.paraformer.cif_predictor import mae_loss
from funasr.train_utils.device_funcs import force_gatherable
from funasr.losses.label_smoothing_loss import LabelSmoothingLoss
from funasr.models.transformer.utils.add_sos_eos import add_sos_eos
from funasr.models.transformer.utils.nets_utils import make_pad_mask, pad_list
from funasr.utils.load_utils import load_audio_text_image_video, extract_fbank


class UniASR(FunASRModel):
@tables.register("model_classes", "UniASR")
class UniASR(torch.nn.Module):
    """
    Author: Speech Lab of DAMO Academy, Alibaba Group
    """

    def __init__(
        self,
        vocab_size: int,
        token_list: Union[Tuple[str, ...], List[str]],
        frontend: Optional[AbsFrontend],
        specaug: Optional[AbsSpecAug],
        normalize: Optional[AbsNormalize],
        encoder: AbsEncoder,
        decoder: AbsDecoder,
        ctc: CTC,
        specaug: Optional[str] = None,
        specaug_conf: Optional[Dict] = None,
        normalize: str = None,
        normalize_conf: Optional[Dict] = None,
        encoder: str = None,
        encoder_conf: Optional[Dict] = None,
        decoder: str = None,
        decoder_conf: Optional[Dict] = None,
        ctc: str = None,
        ctc_conf: Optional[Dict] = None,
        predictor: str = None,
        predictor_conf: Optional[Dict] = None,
        ctc_weight: float = 0.5,
        interctc_weight: float = 0.0,
        input_size: int = 80,
        vocab_size: int = -1,
        ignore_id: int = -1,
        blank_id: int = 0,
        sos: int = 1,
        eos: int = 2,
        lsm_weight: float = 0.0,
        length_normalized_loss: bool = False,
        report_cer: bool = True,
        report_wer: bool = True,
        sym_space: str = "<space>",
        sym_blank: str = "<blank>",
        extract_feats_in_collect_stats: bool = True,
        predictor=None,
        # report_cer: bool = True,
        # report_wer: bool = True,
        # sym_space: str = "<space>",
        # sym_blank: str = "<blank>",
        # extract_feats_in_collect_stats: bool = True,
        # predictor=None,
        predictor_weight: float = 0.0,
        decoder_attention_chunk_type: str = 'chunk',
        encoder2: AbsEncoder = None,
        decoder2: AbsDecoder = None,
        ctc2: CTC = None,
        ctc_weight2: float = 0.5,
        interctc_weight2: float = 0.0,
        predictor2=None,
        predictor_weight2: float = 0.0,
        decoder_attention_chunk_type2: str = 'chunk',
        stride_conv=None,
        loss_weight_model1: float = 0.5,
        enable_maas_finetune: bool = False,
        freeze_encoder2: bool = False,
        preencoder: Optional[AbsPreEncoder] = None,
        postencoder: Optional[AbsPostEncoder] = None,
        predictor_bias: int = 0,
        sampling_ratio: float = 0.2,
        share_embedding: bool = False,
        # preencoder: Optional[AbsPreEncoder] = None,
        # postencoder: Optional[AbsPostEncoder] = None,
        use_1st_decoder_loss: bool = False,
        encoder1_encoder2_joint_training: bool = True,
        **kwargs,
        
    ):
        assert 0.0 <= ctc_weight <= 1.0, ctc_weight
        assert 0.0 <= interctc_weight < 1.0, interctc_weight
@@ -443,10 +431,8 @@
        # force_gatherable: to-device and to-tensor if scalar for DataParallel
        if self.length_normalized_loss:
            batch_size = int((text_lengths + 1).sum())
<<<<<<< HEAD:funasr/models/uniasr/e2e_uni_asr.py

=======
>>>>>>> main:funasr/models/e2e_uni_asr.py

        loss, stats, weight = force_gatherable((loss, stats, batch_size), loss.device)
        return loss, stats, weight


 funasr/models/uniasr/template.yaml

New file
@@ -0,0 +1,178 @@
# This is an example that demonstrates how to configure a model file.
# You can modify the configuration according to your own requirements.

# to print the register_table:
# from funasr.register import tables
# tables.print()

# network architecture
model: UniASR
model_conf:
    ctc_weight: 0.0
    lsm_weight: 0.1
    length_normalized_loss: true
    predictor_weight: 1.0
    decoder_attention_chunk_type: chunk
    ctc_weight2: 0.0
    predictor_weight2: 1.0
    decoder_attention_chunk_type2: chunk
    loss_weight_model1: 0.5

# encoder
encoder: SANMEncoderChunkOpt
encoder_conf:
    output_size: 320
    attention_heads: 4
    linear_units: 1280
    num_blocks: 35
    dropout_rate: 0.1
    positional_dropout_rate: 0.1
    attention_dropout_rate: 0.1
    input_layer: pe
    pos_enc_class: SinusoidalPositionEncoder
    normalize_before: true
    kernel_size: 11
    sanm_shfit: 0
    selfattention_layer_type: sanm
    chunk_size: [20, 60]
    stride: [10, 40]
    pad_left: [5, 10]
    encoder_att_look_back_factor: [0, 0]
    decoder_att_look_back_factor: [0, 0]

# decoder
decoder: FsmnDecoderSCAMAOpt
decoder_conf:
    attention_dim: 256
    attention_heads: 4
    linear_units: 1024
    num_blocks: 12
    dropout_rate: 0.1
    positional_dropout_rate: 0.1
    self_attention_dropout_rate: 0.1
    src_attention_dropout_rate: 0.1
    att_layer_num: 6
    kernel_size: 11
    concat_embeds: true

predictor: CifPredictorV2
predictor_conf:
    idim: 320
    threshold: 1.0
    l_order: 1
    r_order: 1

encoder2: SANMEncoderChunkOpt
encoder2_conf:
    output_size: 320
    attention_heads: 4
    linear_units: 1280
    num_blocks: 20
    dropout_rate: 0.1
    positional_dropout_rate: 0.1
    attention_dropout_rate: 0.1
    input_layer: pe
    pos_enc_class: SinusoidalPositionEncoder
    normalize_before: true
    kernel_size: 21
    sanm_shfit: 0
    selfattention_layer_type: sanm
    chunk_size: [45, 70]
    stride: [35, 50]
    pad_left: [5, 10]
    encoder_att_look_back_factor: [0, 0]
    decoder_att_look_back_factor: [0, 0]

decoder2: FsmnDecoderSCAMAOpt
decoder2_conf:
    attention_dim: 320
    attention_heads: 4
    linear_units: 1280
    num_blocks: 12
    dropout_rate: 0.1
    positional_dropout_rate: 0.1
    self_attention_dropout_rate: 0.1
    src_attention_dropout_rate: 0.1
    att_layer_num: 6
    kernel_size: 11
    concat_embeds: true

predictor2: CifPredictorV2
predictor2_conf:
    idim: 320
    threshold: 1.0
    l_order: 1
    r_order: 1

stride_conv: stride_conv1d
stride_conv_conf:
    kernel_size: 2
    stride: 2
    pad: [0, 1]

# frontend related
frontend: WavFrontendOnline
frontend_conf:
    fs: 16000
    window: hamming
    n_mels: 80
    frame_length: 25
    frame_shift: 10
    lfr_m: 7
    lfr_n: 6

specaug: SpecAugLFR
specaug_conf:
    apply_time_warp: false
    time_warp_window: 5
    time_warp_mode: bicubic
    apply_freq_mask: true
    freq_mask_width_range:
    - 0
    - 30
    lfr_rate: 6
    num_freq_mask: 1
    apply_time_mask: true
    time_mask_width_range:
    - 0
    - 12
    num_time_mask: 1

train_conf:
  accum_grad: 1
  grad_clip: 5
  max_epoch: 150
  keep_nbest_models: 10
  avg_nbest_model: 5
  log_interval: 50

optim: adam
optim_conf:
   lr: 0.0001
scheduler: warmuplr
scheduler_conf:
   warmup_steps: 30000

dataset: AudioDataset
dataset_conf:
    index_ds: IndexDSJsonl
    batch_sampler: DynamicBatchLocalShuffleSampler
    batch_type: example # example or length
    batch_size: 1 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
    max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
    buffer_size: 500
    shuffle: True
    num_workers: 0

tokenizer: CharTokenizer
tokenizer_conf:
  unk_symbol: <unk>
  split_with_space: true


ctc_conf:
    dropout_rate: 0.0
    ctc_type: builtin
    reduce: true
    ignore_nan_grad: true
normalize: null