Remove all-in-one for onnx export

.github/scripts/compute-fbank-librispeech-test-clean-and-test-other.sh

@@ -4,6 +4,8 @@
 # The computed features are saved to ~/tmp/fbank-libri and are
 # cached for later runs
 
+set -e
+
 export PYTHONPATH=$PWD:$PYTHONPATH
 echo $PYTHONPATH
 

.github/scripts/download-gigaspeech-dev-test-dataset.sh

@@ -6,6 +6,8 @@
 # You will find directories `~/tmp/giga-dev-dataset-fbank` after running
 # this script.
 
+set -e
+
 mkdir -p ~/tmp
 cd ~/tmp
 

.github/scripts/download-librispeech-test-clean-and-test-other-dataset.sh

@@ -7,6 +7,8 @@
 # You will find directories ~/tmp/download/LibriSpeech after running
 # this script.
 
+set -e
+
 mkdir ~/tmp/download
 cd egs/librispeech/ASR
 ln -s ~/tmp/download .

.github/scripts/install-kaldifeat.sh

@@ -3,6 +3,8 @@
 # This script installs kaldifeat into the directory ~/tmp/kaldifeat
 # which is cached by GitHub actions for later runs.
 
+set -e
+
 mkdir -p ~/tmp
 cd ~/tmp
 git clone https://github.com/csukuangfj/kaldifeat

.github/scripts/prepare-librispeech-test-clean-and-test-other-manifests.sh

@@ -4,6 +4,8 @@
 # to egs/librispeech/ASR/download/LibriSpeech and generates manifest
 # files in egs/librispeech/ASR/data/manifests
 
+set -e
+
 cd egs/librispeech/ASR
 [ ! -e download ] && ln -s ~/tmp/download .
 mkdir -p data/manifests

.github/scripts/run-aishell-pruned-transducer-stateless3-2022-06-20.sh

@@ -1,5 +1,7 @@
 #!/usr/bin/env bash
 
+set -e
+
 log() {
   # This function is from espnet
   local fname=${BASH_SOURCE[1]##*/}

.github/scripts/run-gigaspeech-pruned-transducer-stateless2-2022-05-12.sh

@@ -1,5 +1,7 @@
 #!/usr/bin/env bash
 
+set -e
+
 log() {
   # This function is from espnet
   local fname=${BASH_SOURCE[1]##*/}

.github/scripts/run-librispeech-lstm-transducer-stateless2-2022-09-03.yml

@@ -1,4 +1,6 @@
 #!/usr/bin/env bash
+#
+set -e
 
 log() {
   # This function is from espnet

.github/scripts/run-librispeech-pruned-transducer-stateless-2022-03-12.sh

@@ -1,5 +1,7 @@
 #!/usr/bin/env bash
 
+set -e
+
 log() {
   # This function is from espnet
   local fname=${BASH_SOURCE[1]##*/}

.github/scripts/run-librispeech-pruned-transducer-stateless2-2022-04-29.sh

@@ -1,5 +1,7 @@
 #!/usr/bin/env bash
 
+set -e
+
 log() {
   # This function is from espnet
   local fname=${BASH_SOURCE[1]##*/}

.github/scripts/run-librispeech-pruned-transducer-stateless3-2022-04-29.sh

@@ -1,5 +1,7 @@
 #!/usr/bin/env bash
 
+set -e
+
 log() {
   # This function is from espnet
   local fname=${BASH_SOURCE[1]##*/}

.github/scripts/run-librispeech-pruned-transducer-stateless3-2022-05-13.sh

@@ -1,5 +1,7 @@
 #!/usr/bin/env bash
 
+set -e
+
 log() {
   # This function is from espnet
   local fname=${BASH_SOURCE[1]##*/}
@@ -62,15 +64,13 @@ log "Decode with ONNX models"
   --onnx-joiner-encoder-proj-filename $repo/exp/joiner_encoder_proj.onnx \
   --onnx-joiner-decoder-proj-filename $repo/exp/joiner_decoder_proj.onnx
 
-./pruned_transducer_stateless3/onnx_check_all_in_one.py \
-  --jit-filename $repo/exp/cpu_jit.pt \
-  --onnx-all-in-one-filename $repo/exp/all_in_one.onnx
-
 ./pruned_transducer_stateless3/onnx_pretrained.py \
   --bpe-model $repo/data/lang_bpe_500/bpe.model \
   --encoder-model-filename $repo/exp/encoder.onnx \
   --decoder-model-filename $repo/exp/decoder.onnx \
   --joiner-model-filename $repo/exp/joiner.onnx \
+  --joiner-encoder-proj-model-filename $repo/exp/joiner_encoder_proj.onnx \
+  --joiner-decoder-proj-model-filename $repo/exp/joiner_decoder_proj.onnx \
   $repo/test_wavs/1089-134686-0001.wav \
   $repo/test_wavs/1221-135766-0001.wav \
   $repo/test_wavs/1221-135766-0002.wav

.github/scripts/run-librispeech-pruned-transducer-stateless5-2022-05-13.sh

@@ -1,5 +1,7 @@
 #!/usr/bin/env bash
 
+set -e
+
 log() {
   # This function is from espnet
   local fname=${BASH_SOURCE[1]##*/}

.github/scripts/run-librispeech-streaming-pruned-transducer-stateless2-2022-06-26.sh

@@ -1,5 +1,7 @@
 #!/usr/bin/env bash
 
+set -e
+
 log() {
   # This function is from espnet
   local fname=${BASH_SOURCE[1]##*/}

.github/scripts/run-librispeech-transducer-stateless2-2022-04-19.sh

@@ -1,5 +1,7 @@
 #!/usr/bin/env bash
 
+set -e
+
 log() {
   # This function is from espnet
   local fname=${BASH_SOURCE[1]##*/}

.github/scripts/run-pre-trained-conformer-ctc.sh

@@ -1,5 +1,7 @@
 #!/usr/bin/env bash
 
+set -e
+
 log() {
   # This function is from espnet
   local fname=${BASH_SOURCE[1]##*/}

.github/scripts/run-pre-trained-transducer-stateless-librispeech-100h.sh

@@ -1,5 +1,7 @@
 #!/usr/bin/env bash
 
+set -e
+
 log() {
   # This function is from espnet
   local fname=${BASH_SOURCE[1]##*/}

.github/scripts/run-pre-trained-transducer-stateless-librispeech-960h.sh

@@ -1,5 +1,7 @@
 #!/usr/bin/env bash
 
+set -e
+
 log() {
   # This function is from espnet
   local fname=${BASH_SOURCE[1]##*/}

.github/scripts/run-pre-trained-transducer-stateless-modified-2-aishell.sh

@@ -1,5 +1,7 @@
 #!/usr/bin/env bash
 
+set -e
+
 log() {
   # This function is from espnet
   local fname=${BASH_SOURCE[1]##*/}

.github/scripts/run-pre-trained-transducer-stateless-modified-aishell.sh

@@ -1,5 +1,7 @@
 #!/usr/bin/env bash
 
+set -e
+
 log() {
   # This function is from espnet
   local fname=${BASH_SOURCE[1]##*/}

.github/scripts/run-pre-trained-transducer-stateless.sh

@@ -1,5 +1,7 @@
 #!/usr/bin/env bash
 
+set -e
+
 log() {
   # This function is from espnet
   local fname=${BASH_SOURCE[1]##*/}

.github/scripts/run-pre-trained-transducer.sh

@@ -1,5 +1,7 @@
 #!/usr/bin/env bash
 
+set -e
+
 log() {
   # This function is from espnet
   local fname=${BASH_SOURCE[1]##*/}

egs/librispeech/ASR/pruned_transducer_stateless2/conformer.py

@@ -476,8 +476,8 @@ def forward(
         self,
         src: Tensor,
         pos_emb: Tensor,
-        src_mask: Optional[Tensor] = None,
         src_key_padding_mask: Optional[Tensor] = None,
+        src_mask: Optional[Tensor] = None,
         warmup: float = 1.0,
     ) -> Tensor:
         """
@@ -486,8 +486,8 @@ def forward(
         Args:
             src: the sequence to the encoder layer (required).
             pos_emb: Positional embedding tensor (required).
-            src_mask: the mask for the src sequence (optional).
             src_key_padding_mask: the mask for the src keys per batch (optional).
+            src_mask: the mask for the src sequence (optional).
             warmup: controls selective bypass of of layers; if < 1.0, we will
               bypass layers more frequently.
         Shape:
@@ -663,17 +663,17 @@ def forward(
         self,
         src: Tensor,
         pos_emb: Tensor,
-        mask: Optional[Tensor] = None,
         src_key_padding_mask: Optional[Tensor] = None,
+        mask: Optional[Tensor] = None,
         warmup: float = 1.0,
     ) -> Tensor:
         r"""Pass the input through the encoder layers in turn.
 
         Args:
             src: the sequence to the encoder (required).
             pos_emb: Positional embedding tensor (required).
-            mask: the mask for the src sequence (optional).
             src_key_padding_mask: the mask for the src keys per batch (optional).
+            mask: the mask for the src sequence (optional).
             warmup: controls selective bypass of of layers; if < 1.0, we will
               bypass layers more frequently.