Add streaming HLG decoding for zipformer CTC.

k2-fsa · Mar 18, 2024 · 40668e4 · 40668e4
1 parent eec12f0
commit 40668e4
Show file tree

Hide file tree

Showing 3 changed files with 443 additions and 6 deletions.
diff --git a/egs/librispeech/ASR/zipformer/export-onnx-streaming-ctc.py b/egs/librispeech/ASR/zipformer/export-onnx-streaming-ctc.py
@@ -32,7 +32,7 @@
   --joiner-dim 512 \
   --causal True \
   --chunk-size 16 \
-  --left-context-frames 64 \
+  --left-context-frames 128 \
   --use-ctc 1
 
 The --chunk-size in training is "16,32,64,-1", so we select one of them
@@ -41,7 +41,7 @@
 
 It will generate the following file inside $repo/exp:
 
-  - ctc-epoch-99-avg-1-chunk-16-left-64.onnx
+  - ctc-epoch-99-avg-1-chunk-16-left-128.onnx
 
 See ./onnx_pretrained-streaming-ctc.py for how to use the exported ONNX models.
 """

diff --git a/egs/librispeech/ASR/zipformer/export-onnx-streaming.py b/egs/librispeech/ASR/zipformer/export-onnx-streaming.py
@@ -48,17 +48,17 @@
   --joiner-dim 512 \
   --causal True \
   --chunk-size 16 \
-  --left-context-frames 64
+  --left-context-frames 128
 
 The --chunk-size in training is "16,32,64,-1", so we select one of them
 (excluding -1) during streaming export. The same applies to `--left-context`,
 whose value is "64,128,256,-1".
 
 It will generate the following 3 files inside $repo/exp:
 
-  - encoder-epoch-99-avg-1-chunk-16-left-64.onnx
-  - decoder-epoch-99-avg-1-chunk-16-left-64.onnx
-  - joiner-epoch-99-avg-1-chunk-16-left-64.onnx
+  - encoder-epoch-99-avg-1-chunk-16-left-128.onnx
+  - decoder-epoch-99-avg-1-chunk-16-left-128.onnx
+  - joiner-epoch-99-avg-1-chunk-16-left-128.onnx
 
 See ./onnx_pretrained-streaming.py for how to use the exported ONNX models.
 """