torchaudio.prototype.models.conformer_rnnt_model¶
- torchaudio.prototype.models.conformer_rnnt_model(*, input_dim: int, encoding_dim: int, time_reduction_stride: int, conformer_input_dim: int, conformer_ffn_dim: int, conformer_num_layers: int, conformer_num_heads: int, conformer_depthwise_conv_kernel_size: int, conformer_dropout: float, num_symbols: int, symbol_embedding_dim: int, num_lstm_layers: int, lstm_hidden_dim: int, lstm_layer_norm: bool, lstm_layer_norm_epsilon: float, lstm_dropout: float, joiner_activation: str) RNNT[源代码]¶
构建基于 Conformer 的循环神经网络 transducer (RNN-T) 模型。
- 参数：
input_dim (int) – 传递给转录网络的输入序列帧的维度。
encoding_dim (int) – 传递给联合网络的由转录网络和预测网络生成的编码维度。
time_reduction_stride (int) – 减小输入序列长度的因子。
conformer_input_dim (int) – Conformer 输入的维度。
conformer_ffn_dim (int) – 每个 Conformer 层的全连接网络隐藏层维度。
conformer_num_layers (int) – 要实例化的 Conformer 层数。
conformer_num_heads (int) – 每个 Conformer 层中的注意力头数量。
conformer_depthwise_conv_kernel_size (int) – 每个 Conformer 层的深度可分离卷积层的核大小。
conformer_dropout (float) – Conformer dropout 概率。
num_symbols (int) – 目标 token 集合的基数。
symbol_embedding_dim (int) – 每个目标 token 嵌入的维度。
num_lstm_layers (int) – 要实例化的 LSTM 层数。
lstm_hidden_dim (int) – 每个 LSTM 层的输出维度。
lstm_layer_norm (bool) – 如果为
True，则为 LSTM 层启用层归一化。lstm_layer_norm_epsilon (float) – 在 LSTM 层归一化层中使用的 epsilon 值。
lstm_dropout (float) – LSTM dropout 概率。
joiner_activation (str) – 联合器中使用的激活函数。必须是 (“relu”, “tanh”) 之一。(默认: “relu”)
返回：
- RNNT
Conformer RNN-T 模型。
