手把手教你语音识别（三）

朋友们，手把手语音识别第三部分来了，这部分开始讲解网络搭建部分，同样是手把手教你哦，不要错过。

1、读入数据

这部分，就不展开讲了，之前的文章讲过：数据处理，这次直接从读入数据的格式到怼入网络开始讲，不过，由于网络很大，层数很多，所以要一层层的去讲解，本次咱们讲解的是encode里面的embeding层，也就是词嵌入层，这里可以认为embedding就是一个lookup，也就是一个速查表，用id->embedding做一个对应，这样的话，根据词的id，能通过vocab.txt查找到这个id对应的字或词，比如a、b、c、d、你、我等，也可以找到这个词的embedding，从而计算他们的关系，比如距离、语义关系等等。

import soundfile
audio, audio_sample_rate = soundfile.read("C:\Users\Desktop\asr16.wav", dtype="int16",always_2d=True)
1
2

import numpy as np
audio = audio.mean(axis=1, dtype=np.int16)
1
2

def pcm16to32(audio):
    assert (audio.dtype == np.int16)
    audio = audio.astype("float32")
    bits = np.iinfo(np.int16).bits
    audio = audio / (2**(bits - 1))
    return audio
def pcm32to16(audio):
    assert (audio.dtype == np.float32)
    bits = np.iinfo(np.int16).bits
    audio = audio * (2**(bits - 1))
    audio = np.round(audio).astype("int16")
    return audio
1
2
3
4
5
6
7
8
9
10
11
12

audio = pcm16to32(audio)
audio=pcm32to16(audio)
1
2

with open("transform.pkl", "rb") as tf:    
    preprocessing = pickle.load(tf)
1
2

audio = preprocessing(audio,**{"train":False}
1

audio = paddle.to_tensor(audio, dtype='float32').unsqueeze(axis=0)
1

2、进入模型

前面的数据读取已经完成，并且进行了transform，也就是预处理，经过这些操作以后，特征已经经过预处理，就比较好用了，具体经过那些处理，前面的文章讲过了，忘记的可以自行翻阅，下面就是要进入网络层，进入网络层之前，可以看到，现在的audio的shape为(1,363,80)，前面的文章也讲过，这个363很重要，需要记住这个数，从现在开始就可以忘记了，因为后面经过很多次reshape后，这个363就没了。

（1）查看网络

在搭建网络之前，得先看看网络啥样，咱们先展示下网络层把，这个之前加载网络讲过，这里展示下网络的summary。

U2Model(
  (encoder): ConformerEncoder(
    (embed): Conv2dSubsampling4(
      (pos_enc): RelPositionalEncoding(
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
      )
      (conv): Sequential(
        (0): Conv2D(1, 512, kernel_size=[3, 3], stride=[2, 2], data_format=NCHW)
        (1): ReLU()
        (2): Conv2D(512, 512, kernel_size=[3, 3], stride=[2, 2], data_format=NCHW)
        (3): ReLU()
      )
      (out): Sequential(
        (0): Linear(in_features=9728, out_features=512, dtype=float32)
      )
    )
    (after_norm): LayerNorm(normalized_shape=[512], epsilon=1e-12)
    (encoders): LayerList(
      (0): ConformerEncoderLayer(
        (self_attn): RelPositionMultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
          (linear_pos): Linear(in_features=512, out_features=512, dtype=float32)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (feed_forward_macaron): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (conv_module): ConvolutionModule(
          (pointwise_conv1): Conv1D(512, 1024, kernel_size=[1], data_format=NCL)
          (depthwise_conv): Conv1D(512, 512, kernel_size=[15], padding=7, groups=512, data_format=NCL)
          (norm): LayerNorm(normalized_shape=[512], epsilon=1e-05)
          (pointwise_conv2): Conv1D(512, 512, kernel_size=[1], data_format=NCL)
          (activation): Swish()
        )
        (norm_ff): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_mha): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_ff_macaron): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_conv): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_final): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear): Linear(in_features=1024, out_features=512, dtype=float32)
      )
      (1): ConformerEncoderLayer(
        (self_attn): RelPositionMultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
          (linear_pos): Linear(in_features=512, out_features=512, dtype=float32)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (feed_forward_macaron): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (conv_module): ConvolutionModule(
          (pointwise_conv1): Conv1D(512, 1024, kernel_size=[1], data_format=NCL)
          (depthwise_conv): Conv1D(512, 512, kernel_size=[15], padding=7, groups=512, data_format=NCL)
          (norm): LayerNorm(normalized_shape=[512], epsilon=1e-05)
          (pointwise_conv2): Conv1D(512, 512, kernel_size=[1], data_format=NCL)
          (activation): Swish()
        )
        (norm_ff): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_mha): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_ff_macaron): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_conv): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_final): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear): Linear(in_features=1024, out_features=512, dtype=float32)
      )
      (2): ConformerEncoderLayer(
        (self_attn): RelPositionMultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
          (linear_pos): Linear(in_features=512, out_features=512, dtype=float32)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (feed_forward_macaron): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (conv_module): ConvolutionModule(
          (pointwise_conv1): Conv1D(512, 1024, kernel_size=[1], data_format=NCL)
          (depthwise_conv): Conv1D(512, 512, kernel_size=[15], padding=7, groups=512, data_format=NCL)
          (norm): LayerNorm(normalized_shape=[512], epsilon=1e-05)
          (pointwise_conv2): Conv1D(512, 512, kernel_size=[1], data_format=NCL)
          (activation): Swish()
        )
        (norm_ff): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_mha): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_ff_macaron): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_conv): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_final): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear): Linear(in_features=1024, out_features=512, dtype=float32)
      )
      (3): ConformerEncoderLayer(
        (self_attn): RelPositionMultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
          (linear_pos): Linear(in_features=512, out_features=512, dtype=float32)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (feed_forward_macaron): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (conv_module): ConvolutionModule(
          (pointwise_conv1): Conv1D(512, 1024, kernel_size=[1], data_format=NCL)
          (depthwise_conv): Conv1D(512, 512, kernel_size=[15], padding=7, groups=512, data_format=NCL)
          (norm): LayerNorm(normalized_shape=[512], epsilon=1e-05)
          (pointwise_conv2): Conv1D(512, 512, kernel_size=[1], data_format=NCL)
          (activation): Swish()
        )
        (norm_ff): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_mha): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_ff_macaron): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_conv): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_final): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear): Linear(in_features=1024, out_features=512, dtype=float32)
      )
      (4): ConformerEncoderLayer(
        (self_attn): RelPositionMultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
          (linear_pos): Linear(in_features=512, out_features=512, dtype=float32)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (feed_forward_macaron): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (conv_module): ConvolutionModule(
          (pointwise_conv1): Conv1D(512, 1024, kernel_size=[1], data_format=NCL)
          (depthwise_conv): Conv1D(512, 512, kernel_size=[15], padding=7, groups=512, data_format=NCL)
          (norm): LayerNorm(normalized_shape=[512], epsilon=1e-05)
          (pointwise_conv2): Conv1D(512, 512, kernel_size=[1], data_format=NCL)
          (activation): Swish()
        )
        (norm_ff): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_mha): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_ff_macaron): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_conv): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_final): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear): Linear(in_features=1024, out_features=512, dtype=float32)
      )
      (5): ConformerEncoderLayer(
        (self_attn): RelPositionMultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
          (linear_pos): Linear(in_features=512, out_features=512, dtype=float32)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (feed_forward_macaron): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (conv_module): ConvolutionModule(
          (pointwise_conv1): Conv1D(512, 1024, kernel_size=[1], data_format=NCL)
          (depthwise_conv): Conv1D(512, 512, kernel_size=[15], padding=7, groups=512, data_format=NCL)
          (norm): LayerNorm(normalized_shape=[512], epsilon=1e-05)
          (pointwise_conv2): Conv1D(512, 512, kernel_size=[1], data_format=NCL)
          (activation): Swish()
        )
        (norm_ff): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_mha): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_ff_macaron): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_conv): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_final): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear): Linear(in_features=1024, out_features=512, dtype=float32)
      )
      (6): ConformerEncoderLayer(
        (self_attn): RelPositionMultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
          (linear_pos): Linear(in_features=512, out_features=512, dtype=float32)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (feed_forward_macaron): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (conv_module): ConvolutionModule(
          (pointwise_conv1): Conv1D(512, 1024, kernel_size=[1], data_format=NCL)
          (depthwise_conv): Conv1D(512, 512, kernel_size=[15], padding=7, groups=512, data_format=NCL)
          (norm): LayerNorm(normalized_shape=[512], epsilon=1e-05)
          (pointwise_conv2): Conv1D(512, 512, kernel_size=[1], data_format=NCL)
          (activation): Swish()
        )
        (norm_ff): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_mha): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_ff_macaron): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_conv): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_final): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear): Linear(in_features=1024, out_features=512, dtype=float32)
      )
      (7): ConformerEncoderLayer(
        (self_attn): RelPositionMultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
          (linear_pos): Linear(in_features=512, out_features=512, dtype=float32)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (feed_forward_macaron): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (conv_module): ConvolutionModule(
          (pointwise_conv1): Conv1D(512, 1024, kernel_size=[1], data_format=NCL)
          (depthwise_conv): Conv1D(512, 512, kernel_size=[15], padding=7, groups=512, data_format=NCL)
          (norm): LayerNorm(normalized_shape=[512], epsilon=1e-05)
          (pointwise_conv2): Conv1D(512, 512, kernel_size=[1], data_format=NCL)
          (activation): Swish()
        )
        (norm_ff): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_mha): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_ff_macaron): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_conv): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_final): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear): Linear(in_features=1024, out_features=512, dtype=float32)
      )
      (8): ConformerEncoderLayer(
        (self_attn): RelPositionMultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
          (linear_pos): Linear(in_features=512, out_features=512, dtype=float32)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (feed_forward_macaron): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (conv_module): ConvolutionModule(
          (pointwise_conv1): Conv1D(512, 1024, kernel_size=[1], data_format=NCL)
          (depthwise_conv): Conv1D(512, 512, kernel_size=[15], padding=7, groups=512, data_format=NCL)
          (norm): LayerNorm(normalized_shape=[512], epsilon=1e-05)
          (pointwise_conv2): Conv1D(512, 512, kernel_size=[1], data_format=NCL)
          (activation): Swish()
        )
        (norm_ff): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_mha): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_ff_macaron): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_conv): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_final): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear): Linear(in_features=1024, out_features=512, dtype=float32)
      )
      (9): ConformerEncoderLayer(
        (self_attn): RelPositionMultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
          (linear_pos): Linear(in_features=512, out_features=512, dtype=float32)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (feed_forward_macaron): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (conv_module): ConvolutionModule(
          (pointwise_conv1): Conv1D(512, 1024, kernel_size=[1], data_format=NCL)
          (depthwise_conv): Conv1D(512, 512, kernel_size=[15], padding=7, groups=512, data_format=NCL)
          (norm): LayerNorm(normalized_shape=[512], epsilon=1e-05)
          (pointwise_conv2): Conv1D(512, 512, kernel_size=[1], data_format=NCL)
          (activation): Swish()
        )
        (norm_ff): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_mha): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_ff_macaron): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_conv): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_final): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear): Linear(in_features=1024, out_features=512, dtype=float32)
      )
      (10): ConformerEncoderLayer(
        (self_attn): RelPositionMultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
          (linear_pos): Linear(in_features=512, out_features=512, dtype=float32)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (feed_forward_macaron): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (conv_module): ConvolutionModule(
          (pointwise_conv1): Conv1D(512, 1024, kernel_size=[1], data_format=NCL)
          (depthwise_conv): Conv1D(512, 512, kernel_size=[15], padding=7, groups=512, data_format=NCL)
          (norm): LayerNorm(normalized_shape=[512], epsilon=1e-05)
          (pointwise_conv2): Conv1D(512, 512, kernel_size=[1], data_format=NCL)
          (activation): Swish()
        )
        (norm_ff): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_mha): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_ff_macaron): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_conv): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_final): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear): Linear(in_features=1024, out_features=512, dtype=float32)
      )
      (11): ConformerEncoderLayer(
        (self_attn): RelPositionMultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
          (linear_pos): Linear(in_features=512, out_features=512, dtype=float32)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (feed_forward_macaron): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): Swish()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (conv_module): ConvolutionModule(
          (pointwise_conv1): Conv1D(512, 1024, kernel_size=[1], data_format=NCL)
          (depthwise_conv): Conv1D(512, 512, kernel_size=[15], padding=7, groups=512, data_format=NCL)
          (norm): LayerNorm(normalized_shape=[512], epsilon=1e-05)
          (pointwise_conv2): Conv1D(512, 512, kernel_size=[1], data_format=NCL)
          (activation): Swish()
        )
        (norm_ff): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_mha): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_ff_macaron): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_conv): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm_final): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear): Linear(in_features=1024, out_features=512, dtype=float32)
      )
    )
  )
  (decoder): TransformerDecoder(
    (embed): Sequential(
      (0): Embedding(5537, 512, sparse=False)
      (1): PositionalEncoding(
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
      )
    )
    (after_norm): LayerNorm(normalized_shape=[512], epsilon=1e-12)
    (output_layer): Linear(in_features=512, out_features=5537, dtype=float32)
    (decoders): LayerList(
      (0): DecoderLayer(
        (self_attn): MultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
        )
        (src_attn): MultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): ReLU()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (norm1): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm2): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm3): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear1): Linear(in_features=1024, out_features=512, dtype=float32)
        (concat_linear2): Linear(in_features=1024, out_features=512, dtype=float32)
      )
      (1): DecoderLayer(
        (self_attn): MultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
        )
        (src_attn): MultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): ReLU()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (norm1): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm2): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm3): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear1): Linear(in_features=1024, out_features=512, dtype=float32)
        (concat_linear2): Linear(in_features=1024, out_features=512, dtype=float32)
      )
      (2): DecoderLayer(
        (self_attn): MultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
        )
        (src_attn): MultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): ReLU()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (norm1): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm2): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm3): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear1): Linear(in_features=1024, out_features=512, dtype=float32)
        (concat_linear2): Linear(in_features=1024, out_features=512, dtype=float32)
      )
      (3): DecoderLayer(
        (self_attn): MultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
        )
        (src_attn): MultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): ReLU()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (norm1): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm2): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm3): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear1): Linear(in_features=1024, out_features=512, dtype=float32)
        (concat_linear2): Linear(in_features=1024, out_features=512, dtype=float32)
      )
      (4): DecoderLayer(
        (self_attn): MultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
        )
        (src_attn): MultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): ReLU()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (norm1): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm2): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm3): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear1): Linear(in_features=1024, out_features=512, dtype=float32)
        (concat_linear2): Linear(in_features=1024, out_features=512, dtype=float32)
      )
      (5): DecoderLayer(
        (self_attn): MultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
        )
        (src_attn): MultiHeadedAttention(
          (linear_q): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_k): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_v): Linear(in_features=512, out_features=512, dtype=float32)
          (linear_out): Linear(in_features=512, out_features=512, dtype=float32)
          (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
        )
        (feed_forward): PositionwiseFeedForward(
          (w_1): Linear(in_features=512, out_features=2048, dtype=float32)
          (activation): ReLU()
          (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
          (w_2): Linear(in_features=2048, out_features=512, dtype=float32)
        )
        (norm1): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm2): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (norm3): LayerNorm(normalized_shape=[512], epsilon=1e-12)
        (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
        (concat_linear1): Linear(in_features=1024, out_features=512, dtype=float32)
        (concat_linear2): Linear(in_features=1024, out_features=512, dtype=float32)
      )
    )
  )
  (ctc): CTCDecoderBase(
    (dropout): Dropout(p=0.0, axis=None, mode=upscale_in_train)
    (ctc_lo): Linear(in_features=512, out_features=5537, dtype=float32)
    (criterion): CTCLoss(
      (loss): CTCLoss()
    )
  )
  (criterion_att): LabelSmoothingLoss(
    (criterion): KLDivLoss()
  )
)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643

这个网络非常长，所以需要拆开慢慢讲解，下面我们先讲这一层，然后这一层，主要是两部分：一个是卷积，一个是Linear，然后我们来手工实现。

Conv2dSubsampling4(
  (pos_enc): RelPositionalEncoding(
    (dropout): Dropout(p=0.1, axis=None, mode=upscale_in_train)
  )
  (conv): Sequential(
    (0): Conv2D(1, 512, kernel_size=[3, 3], stride=[2, 2], data_format=NCHW)
    (1): ReLU()
    (2): Conv2D(512, 512, kernel_size=[3, 3], stride=[2, 2], data_format=NCHW)
    (3): ReLU()
  )
  (out): Sequential(
    (0): Linear(in_features=9728, out_features=512, dtype=float32)
  )
)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

（2）搭建网络

import paddle
class Model(paddle.nn.Layer):
    def __init__(self):
        super(Model,self).__init__()
        self.conv2d_1 = paddle.nn.Conv2D(1, 512, kernel_size=[3, 3], stride=[2, 2], data_format="NCHW")
        self.conv2d_2 = paddle.nn.Conv2D(512, 512, kernel_size=[3, 3], stride=[2, 2], data_format="NCHW")
        self.lineone = paddle.nn.Linear(9728,512)
    def forward(self,inputs):
        y = self.conv2d_1(inputs)
        y = paddle.nn.functional.relu(y)
        y = self.conv2d_2(y)
        y = paddle.nn.functional.relu(y)
        b, c, t, f = paddle.shape(y)
        y = y.transpose([0, 2, 1, 3]).reshape([b,t,c*f])
        y = self.lineone(y)
        print(y.shape)
        print(y)
        return y
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18

上面这是使用paddle框架，根据最终的summary搭建出来的网络，具体为何如此搭建，请看下图。

可以清楚的看到，在forward过程中，先是卷积，然后卷积完以后，进行了transpose，reshape才怼入Linear，所以咱们也得这么写。

model = Model()
1

audio1 = audio.unsqueeze(1) # (b, c=1, t, f)
1

x = model(audio1)
1

可以看到，这个时候，咱们自己写的网络和源代码里面的都对上了，输出shape也一样。
（3）embedding

这一部分其实就是一个lookup表，不写代码了，直接看源码，不过这个地方主要是position_embed，最终是多个结果一起输出

完成位置嵌入后，其实后面还有很多处理，这个下次再讲了。

3、总结

好了，这里讲了这么多，其实只是完成了网络的第一部分，后面还有很多，下期继续讲。