minor fixes

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

@@ -7,7 +7,7 @@
 # this script.
 
 mkdir -p ~/tmp
-cd tmp
+cd ~/tmp
 
 git lfs install
 git clone https://huggingface.co/csukuangfj/giga-dev-dataset-fbank

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

@@ -0,0 +1,50 @@
+#!/usr/bin/env bash
+
+log() {
+  # This function is from espnet
+  local fname=${BASH_SOURCE[1]##*/}
+  echo -e "$(date '+%Y-%m-%d %H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*"
+}
+
+cd egs/gigaspeech/ASR
+
+repo_url=https://huggingface.co/wgb14/icefall-asr-gigaspeech-pruned-transducer-stateless2
+
+log "Downloading pre-trained model from $repo_url"
+git lfs install
+git clone $repo_url
+repo=$(basename $repo_url)
+
+echo "GITHUB_EVENT_NAME: ${GITHUB_EVENT_NAME}"
+echo "GITHUB_EVENT_LABEL_NAME: ${GITHUB_EVENT_LABEL_NAME}"
+# if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" || x"${GITHUB_EVENT_LABEL_NAME}" == x"run-decode"  ]]; then
+if [[ x"${GITHUB_EVENT_NAME}" == x"push" || x"${GITHUB_EVENT_LABEL_NAME}" == x"run-decode"  ]]; then
+  mkdir -p pruned_transducer_stateless2/exp
+  ln -s $PWD/$repo/exp/pretrained-epoch-29-avg-11.pt pruned_transducer_stateless2/exp/epoch-999.pt
+  ln -s $PWD/$repo/data/lang_bpe_500 data/
+
+  ls -lh data
+  ls -lh data/lang_bpe_500
+  ls -lh data/fbank
+  ls -lh pruned_transducer_stateless2/exp
+
+  log "Decoding dev and test"
+
+  # use a small value for decoding with CPU
+  max_duration=100
+
+  # Test only greedy_search to reduce CI running time
+  # for method in greedy_search fast_beam_search modified_beam_search; do
+  for method in greedy_search; do
+    log "Decoding with $method"
+
+    ./pruned_transducer_stateless2/decode.py \
+      --decoding-method $method \
+      --epoch 999 \
+      --avg 1 \
+      --max-duration $max_duration \
+      --exp-dir pruned_transducer_stateless2/exp
+  done
+
+  rm pruned_transducer_stateless2/exp/*.pt
+fi

.github/workflows/run-gigaspeech-2022-05-13.yml

@@ -81,3 +81,41 @@ jobs:
           sudo apt-get install -y -q git-lfs
 
           .github/scripts/download-gigaspeech-dev-test-dataset.sh
+
+      - name: Inference with pre-trained model
+        shell: bash
+        env:
+          GITHUB_EVENT_NAME: ${{ github.event_name }}
+          GITHUB_EVENT_LABEL_NAME: ${{ github.event.label.name }}
+        run: |
+          ln -s ~/tmp/giga-dev-dataset-fbank/data egs/gigaspeech/ASR/
+
+          ls -lh egs/gigaspeech/ASR/data/fbank
+
+          export PYTHONPATH=$PWD:$PYTHONPATH
+          export PYTHONPATH=~/tmp/kaldifeat/kaldifeat/python:$PYTHONPATH
+          export PYTHONPATH=~/tmp/kaldifeat/build/lib:$PYTHONPATH
+
+          .github/scripts/run-gigaspeech-pruned-transducer-stateless2-2022-05-12.sh
+
+      - name: Display decoding results for pruned_transducer_stateless
+        # if: github.event_name == 'schedule' || github.event.label.name == 'run-decode'
+        shell: bash
+        run: |
+          cd egs/gigaspeech/ASR/
+          tree ./pruned_transducer_stateless2/exp
+
+          sudo apt-get -qq install tree
+
+          cd pruned_transducer_stateless2
+          echo "results for pruned_transducer_stateless2"
+          echo "===greedy search==="
+          find exp/greedy_search -name "log-*" -exec grep -n --color "best for dev" {} + | sort -n -k2
+          find exp/greedy_search -name "log-*" -exec grep -n --color "best for test" {} + | sort -n -k2
+
+      - name: Upload decoding results for gigaspeech pruned_transducer_stateless2
+        uses: actions/upload-artifact@v2
+        # if: github.event_name == 'schedule' || github.event.label.name == 'run-decode'
+        with:
+          name: torch-${{ matrix.torch }}-python-${{ matrix.python-version }}-ubuntu-18.04-cpu-gigaspeech-pruned_transducer_stateless2-2022-05-12
+          path: egs/gigaspeech/ASR/pruned_transducer_stateless2/exp/