Forward input chunk by chunk with chunk_size like a streaming fashion Here we should pay special attention to computation cache

beam search for freeze only support batch=1 Args: samples: the data source to be decoded beam_size: beam size

beam search for freeze only support batch=1 Args: samples: the data source to be decoded beam_size: beam size

beam search for freeze only support batch=1 Args: samples: the data source to be decoded beam_size: beam size

CTC prefix beam search inner implementation reference: https://github.com/wenet-e2e/wenet/blob/main/wenet/transformer/asr_model.py. A

entry point for model inference, do some preparation work

freeze TensorFlow model trained on Python to saved model,

entry point for model inference, do some preparation work

generate DCT matrix */

generate frequency scale warping */

Enqueue batch data

TODO: Need to determine if we can mark the return value or *this as const

TODO: Need to determine if we can mark the return value or *this as const

get loss

get MPC loss Args: logits: MPC output Return:: MPC L1 loss and metrics

get loss used for training

get loss used for training

get loss used for training

get loss

get loss

loss is returned by self.get_loss(), and it is a tensor with none shape float value

Gated Linear Unit. Original paper: https://arxiv.org/abs/1612.08083 Args: x: float Tensor to perform activation. Returns: `x