From ce8fb6fea04f7345aebbdaf0143f054ceaa879c3 Mon Sep 17 00:00:00 2001
From: dohe0342 <kimdohe1070@gmail.com>
Date: Sat, 10 Dec 2022 11:08:49 +0900
Subject: [PATCH] from local

---
 egs/librispeech/ASR/.run_v2.sh.swp            | Bin 12288 -> 12288 bytes
 .../.train.py.swp                             | Bin 61440 -> 61440 bytes
 .../train.py                                  |  64 +++++++++---------
 3 files changed, 33 insertions(+), 31 deletions(-)

diff --git a/egs/librispeech/ASR/.run_v2.sh.swp b/egs/librispeech/ASR/.run_v2.sh.swp
index c1ec03e7c03c47ae111b90faa71ada8d648cd9a9..17cc2732262b1eac16d197a491e91ac662f157dc 100644
GIT binary patch
delta 32
mcmZojXh;xGG6?hZRj|}EU;qLE28Qb|CntxUxx7*Qr9J?c-3q7x

delta 32
mcmZojXh;xGG6?hZRj|}EU;qLE28NhVlakNwxU^CHr9J?asS1|>

diff --git a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.train.py.swp b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.train.py.swp
index 468ac8578e17bcb0f44f92b9c2c94e04a1037aa9..b5713ff2bdb9acdebd51653e38523708c4c18b8b 100644
GIT binary patch
delta 658
zcmYk)Pe{{Y7{Kvoe_XD)KaSXJDQp-uC2CYESd5@4*k);T==VoZf<S|Uz*fZk4Z76T
zr@Go7=~Ps#iwM%COC4$wMRl>LLp3ZzqFa4`(V+)E@AL4!@8RWn-s+04y5f8ABIZfe
z>ajh$oHegV#V;Z;5u4=3r>MwB%wYoE@ZwcOL@|v%#NmezAHpKDn8H!y@Uumvfv<Rm
z8z^HKIs6KVBdZ~i1zg4@455G|68N!E<SEXhjQxmUIVf@+*Wh3V6^vs5o%r1>@)C=<
zgDTG8G!A39*%4`{Sl=M>29I$MS22xKIDj@JA$S)Md5ha9qkuLf;e!_*G@3-}xQ{EC
z!~jzGM#o>{ZkWYw{{+VU59{+@T`uhOq_r{HUTchYHq~|<I_J@);_jfN|Al@Z>ea+p
zM59JlGo>CKwBmZ!*sHgU4qY_5^}W%h>!t15WoG`K9W#6NiM3hh%%IMjoAi;H)qok-
zFJ@l1THAEr#e^QWGMb+9>jkTyyHQ=TLga($x3fA^%xcN*Q^(Hhf<368k8BASncs7`
u|0boaBPW9HwAb9-T4kiCCEz}G@}#yJ$^Xh=KAr0q_egIRJ8JQ<)!9Fau8cPT

delta 534
zcmYk(!E4fC9LMp`uiwvfrkfL*9$i8b40UZ4co~rhv0xUZ#%hP$Zpv0P*oe|UKs@wg
zhde6Lp~D70I&`T)AyFO5t?TeO*UAn%L|u9d9rlIS^Zh;#JP&-oPqk}SyXMxHfDv}S
z2Bs2~j45?2Ndf5&X%M{;X&*08Mhav2IV^2r1M^5Bi3EaZ`=uu8SV0QYm_i7@e9|X8
zM-2s^D!M^2j-yl3A>QH*ns|a7ESUJ<l^R&U8GIj-4zP<Ibnprdtl=ikA%MRg=^NhT
z9a?C5D$;$5tBB(?PN3(OK4KdiD5Hd1NFssXC#BDLhB8VhAdeg_<18%fxo8RvEF+Bx
z7%rBM<rvmA(>|Y$J1^6H%ZS?7XD>Q4`L^MS5V2THc5Qajxpe!H>3*=Z7`eADyET`w
zf6Rsab!O*fxPuQ2+oRdy&|vq6)7Q@9>;>QP3&CjQ%5r&S_2E*a`2S^R^lty@Kg6wG
ADF6Tf

diff --git a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/train.py b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/train.py
index ffc257b56..7d62876a8 100755
--- a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/train.py
+++ b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/train.py
@@ -719,41 +719,44 @@ def compute_loss(
 
         loss = simple_loss_scale * simple_loss + pruned_loss_scale * pruned_loss
     
-    '''
-    # Compute ctc loss
+    info = MetricsTracker()
+    
+    if params.ctc_loss_scale > 0:
+        # Compute ctc loss
 
-    # NOTE: We need `encode_supervisions` to sort sequences with
-    # different duration in decreasing order, required by
-    # `k2.intersect_dense` called in `k2.ctc_loss`
-    with warnings.catch_warnings():
-        warnings.simplefilter("ignore")
-        supervision_segments, token_ids = encode_supervisions(
-            supervisions,
-            subsampling_factor=params.subsampling_factor,
-            token_ids=token_ids,
+        # NOTE: We need `encode_supervisions` to sort sequences with
+        # different duration in decreasing order, required by
+        # `k2.intersect_dense` called in `k2.ctc_loss`
+        with warnings.catch_warnings():
+            warnings.simplefilter("ignore")
+            supervision_segments, token_ids = encode_supervisions(
+                supervisions,
+                subsampling_factor=params.subsampling_factor,
+                token_ids=token_ids,
+            )
+
+        # Works with a BPE model
+        decoding_graph = k2.ctc_graph(token_ids, modified=False, device=device)
+        dense_fsa_vec = k2.DenseFsaVec(
+            ctc_output,
+            supervision_segments,
+            allow_truncate=params.subsampling_factor - 1,
         )
 
-    # Works with a BPE model
-    decoding_graph = k2.ctc_graph(token_ids, modified=False, device=device)
-    dense_fsa_vec = k2.DenseFsaVec(
-        ctc_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="sum",
-        use_double_scores=params.use_double_scores,
-    )
-    assert ctc_loss.requires_grad == is_training
-    loss += params.ctc_loss_scale * ctc_loss
-    '''
+        ctc_loss = k2.ctc_loss(
+            decoding_graph=decoding_graph,
+            dense_fsa_vec=dense_fsa_vec,
+            output_beam=params.beam_size,
+            reduction="sum",
+            use_double_scores=params.use_double_scores,
+        )
+        assert ctc_loss.requires_grad == is_training
+        loss += params.ctc_loss_scale * ctc_loss
+    
+        info["ctc_loss"] = ctc_loss.detach().cpu().item()
+    
     assert loss.requires_grad == is_training
 
-    info = MetricsTracker()
     with warnings.catch_warnings():
         warnings.simplefilter("ignore")
         info["frames"] = (feature_lens // params.subsampling_factor).sum().item()
@@ -762,7 +765,6 @@ def compute_loss(
     info["loss"] = loss.detach().cpu().item()
     info["simple_loss"] = simple_loss.detach().cpu().item()
     info["pruned_loss"] = pruned_loss.detach().cpu().item()
-    #info["ctc_loss"] = ctc_loss.detach().cpu().item()
 
     return loss, info