From 326a55eebc5a769b9be9f4e5bead11756fe0e9a7 Mon Sep 17 00:00:00 2001
From: dohe0342 <kimdohe1070@gmail.com>
Date: Tue, 14 Feb 2023 13:00:21 +0900
Subject: [PATCH] from local

---
 egs/tedlium2/ASR/conformer_ctc3/.train.py.swp | Bin 53248 -> 57344 bytes
 egs/tedlium2/ASR/conformer_ctc3/train.py      |  40 ++++++++++++------
 2 files changed, 28 insertions(+), 12 deletions(-)

diff --git a/egs/tedlium2/ASR/conformer_ctc3/.train.py.swp b/egs/tedlium2/ASR/conformer_ctc3/.train.py.swp
index 38d6a2df6b028ce712a6fd4de3478d546acf7eb8..e29f93389d3b266f3007fd7d760ab6f7460ce5e5 100644
GIT binary patch
delta 646
zcmb8sPe@cz6bA5b=DqwkGtL_YMGAc`CKGeiHVs2X3l+glpbSJvJ*W3(rgRkFQ^*!G
ziqj?p^MDHti*^}8I*Uqz0uyPYP%Ui{_K$WExXGYzY;x~~ANRuH+<W;>=An{#tvq#W
zF0Uxsva3s0wCV~U{;lBu-JM(0PWo4M^`bAL0%ONm(10!pT-!}&i-9DaqZ71O1U#oD
zdQ4f0QX~Cl=QastX^Qf<-KG$I;_dU)Ln?h%fOj-QgH%PkZr~Ly(tsPai11KCUtGX@
z+MxTiK#;A8YT57i<X#%<idQczbHFwwGg&Xc%&;_ArzJgAEK#dLZduLp$ZC?eR#5Iv
zR>{5<k;@6cJV}&EcidNz?@y&n=jza?Gd${C3HQm6bv)drpVx1m3ALLk+w8QBs2Mwj
z!c%(E(ua)^W6%yIQ;s<jb7GOhT5dXSC(_1s;A|=Ijq<q$3%P&r1JCFl^-u@3lTJU&
zfj2ZqHz-6p)zKdgaFfQVnZ9~~2Xu*Asg`y;zzgd4pb0HJc<4tNM@M(4m)fX~0`&dh
jgE^Oap>%dum5&o8g;*QagQSj_r1ZvXn07$M<CT8_s|1tj

delta 286
zcmWm8txE%O0LJm>Ues`G-f*&E*-W~zV)7*k28(}!2O9R<RS^vq2U-_yPtr6v#iB61
zW}1SqWVCp5L3}}+_$T-nJ`Yd)HhWI9<8*D`^62<iR={h&kbL^Z$*5PpDZ8tgJb*Jx
z;{yS6Td#+$37{?;QkCCv;9cJ2K~7~y)}<sLW57UCxs_vClrI;EWm7UTa)4(!k$ox2
zw+-~=LiS|M#td>gKT1mFQO@N*G3H%VY`5K{v;E}oeqdNR$}+Am@!(~io~d%#RA|f`
iBQwi1%(8D*7=#`#hXvLeo?QbM%rr;Csif7|4gUbgm`FMR

diff --git a/egs/tedlium2/ASR/conformer_ctc3/train.py b/egs/tedlium2/ASR/conformer_ctc3/train.py
index ffbe41c7c..0557a7c20 100755
--- a/egs/tedlium2/ASR/conformer_ctc3/train.py
+++ b/egs/tedlium2/ASR/conformer_ctc3/train.py
@@ -543,20 +543,36 @@ def compute_loss(
         #token_ids = convert_texts_into_ids(texts, graph_compiler.sp)
         token_ids = graph_compiler.texts_to_ids(texts)
         decoding_graph = graph_compiler.compile(token_ids)
+        
+        if params.interctc:
+            dense_fsa_vec1 = k2.DenseFsaVec(
+                nnet_output[8],
+                supervision_segments,
+                allow_truncate=params.subsampling_factor - 1,
+            )
 
-        dense_fsa_vec = k2.DenseFsaVec(
-            nnet_output,
-            supervision_segments,
-            allow_truncate=params.subsampling_factor - 1,
-        )
+            ctc_loss = k2.ctc_loss(
+                decoding_graph=decoding_graph,
+                dense_fsa_vec=dense_fsa_vec,
+                output_beam=params.beam_size,
+                reduction=params.reduction,
+                use_double_scores=params.use_double_scores,
+            )
 
-        ctc_loss = k2.ctc_loss(
-            decoding_graph=decoding_graph,
-            dense_fsa_vec=dense_fsa_vec,
-            output_beam=params.beam_size,
-            reduction=params.reduction,
-            use_double_scores=params.use_double_scores,
-        )
+        else:
+            dense_fsa_vec = k2.DenseFsaVec(
+                nnet_output,
+                supervision_segments,
+                allow_truncate=params.subsampling_factor - 1,
+            )
+
+            ctc_loss = k2.ctc_loss(
+                decoding_graph=decoding_graph,
+                dense_fsa_vec=dense_fsa_vec,
+                output_beam=params.beam_size,
+                reduction=params.reduction,
+                use_double_scores=params.use_double_scores,
+            )
 
         if params.att_rate > 0.0:
             with torch.set_grad_enabled(is_training):