Add recipe for fine-tuning Zipformer with adapter (#1512)

k2-fsa · Feb 29, 2024 · 7e2b561 · 7e2b561
1 parent d89f4ea
commit 7e2b561
Show file tree

Hide file tree

Showing 18 changed files with 7,251 additions and 0 deletions.
diff --git a/egs/librispeech/ASR/README.md b/egs/librispeech/ASR/README.md
@@ -35,6 +35,7 @@ The following table lists the differences among them.
 | `lstm_transducer_stateless2` | LSTM | Embedding + Conv1d | Using LSTM with mechanisms in reworked model + gigaspeech (multi-dataset setup) |
 | `lstm_transducer_stateless3` | LSTM | Embedding + Conv1d | Using LSTM with mechanisms in reworked model + gradient filter + delay penalty |
 | `zipformer` | Upgraded Zipformer | Embedding + Conv1d | The latest recipe |
+| `zipformer_adapter` | Upgraded Zipformer | Embedding + Conv1d | It supports domain adaptation of Zipformer using parameter efficient adapters |
 
 The decoder in `transducer_stateless` is modified from the paper
 [Rnn-Transducer with Stateless Prediction Network](https://ieeexplore.ieee.org/document/9054419/).

diff --git a/egs/librispeech/ASR/zipformer_adapter/asr_datamodule.py b/egs/librispeech/ASR/zipformer_adapter/asr_datamodule.py
@@ -0,0 +1 @@
+../tdnn_lstm_ctc/asr_datamodule.py
diff --git a/egs/librispeech/ASR/zipformer_adapter/beam_search.py b/egs/librispeech/ASR/zipformer_adapter/beam_search.py
@@ -0,0 +1 @@
+../pruned_transducer_stateless2/beam_search.py