mirror of https://github.com/k2-fsa/icefall.git synced 2025-08-09 10:02:22 +00:00

History

* Add emformer model.

* Copy files.

* Use Emformer model as RNN-T encoder.

* Support streaming decoding.

* Minor fixes.

* Add RNN-T Emformer for Aishell.

2022-04-02 13:37:39 +08:00

conformer_ctc

Reset seed at the beginning of each epoch. (#221 )

2022-02-21 15:16:39 +08:00

conformer_mmi

Reset seed at the beginning of each epoch. (#221 )

2022-02-21 15:16:39 +08:00

local

Remove duplicate files. (#236 )

2022-03-04 11:56:31 +08:00

tdnn_lstm_ctc

Change for asr_datamodule.py (#241 )

2022-03-14 00:30:58 +08:00

transducer_emformer

Use Emformer as RNN-T encoder. (#278 )

2022-04-02 13:37:39 +08:00

transducer_stateless

Add modified transducer loss for AIShell dataset (#219 )

2022-03-02 16:02:38 +08:00

transducer_stateless_modified

Batch decoding for models trained with optimized_transducer (#267 )

2022-03-23 19:11:34 +08:00

transducer_stateless_modified-2

Batch decoding for models trained with optimized_transducer (#267 )

2022-03-23 19:11:34 +08:00

prepare_aidatatang_200zh.sh

Add modified transducer loss for AIShell dataset (#219 )

2022-03-02 16:02:38 +08:00

prepare.sh

Minor fixes for aishell (#218 )

2022-02-19 22:28:19 +08:00

README.md

Add stateless transducer tutorial. (#235 )

2022-03-03 22:33:47 +08:00

RESULTS.md

Add modified transducer loss for AIShell dataset (#219 )

2022-03-02 16:02:38 +08:00

shared

Add aishell recipe (#30 )

2021-11-18 10:00:47 +08:00

README.md

Introduction

Please refer to https://icefall.readthedocs.io/en/latest/recipes/aishell/index.html for how to run models in this recipe.

Transducers

There are various folders containing the name transducer in this folder. The following table lists the differences among them.

	Encoder	Decoder	Comment
`transducer_stateless`	Conformer	Embedding + Conv1d	with `k2.rnnt_loss`
`transducer_stateless_modified`	Conformer	Embedding + Conv1d	with modified transducer from `optimized_transducer`
`transducer_stateless_modified-2`	Conformer	Embedding + Conv1d	with modified transducer from `optimized_transducer` + extra data

The decoder in transducer_stateless is modified from the paper Rnn-Transducer with Stateless Prediction Network. We place an additional Conv1d layer right after the input embedding layer.