Run decode.py in GitHub actions.

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

@@ -0,0 +1,17 @@
+#!/usr/bin/env bash
+
+# This script computes fbank features for the test-clean and test-other datasets.
+# The computed features are saved to ~/tmp/fbank-libri and are
+# cached for later runs
+
+export PYTHONPATH=$PWD:$PYTHONPATH
+echo $PYTHONPATH
+
+mkdir ~/tmp/fbank-libri
+cd egs/librispeech/ASR
+mkdir -p data
+cd data
+[ ! -e fbank ] && ln -s ~/tmp/fbank-libri fbank
+cd ..
+./local/compute_fbank_librispeech.py
+ls -lh data/fbank/

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

@@ -0,0 +1,23 @@
+#!/usr/bin/env bash
+
+# This script downloads the test-clean and test-other datasets
+# of LibriSpeech and unzip them to the folder ~/tmp/download,
+# which is cached by GitHub actions for later runs.
+#
+# You will find directories ~/tmp/download/LibriSpeech after running
+# this script.
+
+mkdir ~/tmp/download
+cd egs/librispeech/ASR
+ln -s ~/tmp/download .
+cd download
+wget -q --no-check-certificate https://www.openslr.org/resources/12/test-clean.tar.gz
+tar xf test-clean.tar.gz
+rm test-clean.tar.gz
+
+wget -q --no-check-certificate https://www.openslr.org/resources/12/test-other.tar.gz
+tar xf test-other.tar.gz
+rm test-other.tar.gz
+pwd
+ls -lh
+ls -lh LibriSpeech

.github/scripts/install-kaldifeat.sh

@@ -0,0 +1,13 @@
+#!/usr/bin/env bash
+
+# This script installs kaldifeat into the directory ~/tmp/kaldifeat
+# which is cached by GitHub actions for later runs.
+
+mkdir -p ~/tmp
+cd ~/tmp
+git clone https://github.com/csukuangfj/kaldifeat
+cd kaldifeat
+mkdir build
+cd build
+cmake -DCMAKE_BUILD_TYPE=Release ..
+make -j2 _kaldifeat

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

@@ -0,0 +1,11 @@
+#!/usr/bin/env bash
+
+# This script assumes that test-clean and test-other are downloaded
+# to egs/librispeech/ASR/download/LibriSpeech and generates manifest
+# files in egs/librispeech/ASR/data/manifests
+
+cd egs/librispeech/ASR
+[ ! -e download ] && ln -s ~/tmp/download .
+mkdir -p data/manifests
+lhotse prepare librispeech -j 2 -p test-clean -p test-other ./download/LibriSpeech data/manifests
+ls -lh data/manifests

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

@@ -45,3 +45,31 @@ for method in modified_beam_search beam_search; do
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav
 done
+
+echo "GITHUB_EVENT_NAME: ${GITHUB_EVENT_NAME}"
+if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" ]]; then
+  mkdir -p pruned_transducer_stateless/exp
+  ln -s $PWD/$repo/exp/pretrained.pt pruned_transducer_stateless/exp/epoch-999.pt
+  ln -s $PWD/$repo/data/lang_bpe_500 data/
+
+  ls -lh data
+  ls -lh pruned_transducer_stateless/exp
+
+  log "Decoding test-clean and test-other"
+
+  # use a small value for decoding with CPU
+  max_duration=50
+
+  for method in greedy_search fast_beam_search; do
+    log "Decoding with $method"
+
+    ./pruned_transducer_stateless/decode.py \
+      --decoding-method $method \
+      --epoch 999 \
+      --avg 1 \
+      --max-duration $max_duration \
+      --exp-dir pruned_transducer_stateless/exp
+  done
+
+  rm pruned_transducer_stateless/exp/*.pt
+fi

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

@@ -49,3 +49,31 @@ for method in modified_beam_search beam_search fast_beam_search; do
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav
 done
+
+echo "GITHUB_EVENT_NAME: ${GITHUB_EVENT_NAME}"
+if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" ]]; then
+  mkdir -p pruned_transducer_stateless2/exp
+  ln -s $PWD/$repo/exp/pretrained.pt pruned_transducer_stateless2/exp/epoch-999.pt
+  ln -s $PWD/$repo/data/lang_bpe_500 data/
+
+  ls -lh data
+  ls -lh pruned_transducer_stateless2/exp
+
+  log "Decoding test-clean and test-other"
+
+  # use a small value for decoding with CPU
+  max_duration=50
+
+  for method in greedy_search fast_beam_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/scripts/run-librispeech-pruned-transducer-stateless3-2022-04-29.sh

@@ -49,3 +49,31 @@ for method in modified_beam_search beam_search fast_beam_search; do
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav
 done
+
+echo "GITHUB_EVENT_NAME: ${GITHUB_EVENT_NAME}"
+if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" ]]; then
+  mkdir -p pruned_transducer_stateless3/exp
+  ln -s $PWD/$repo/exp/pretrained.pt pruned_transducer_stateless3/exp/epoch-999.pt
+  ln -s $PWD/$repo/data/lang_bpe_500 data/
+
+  ls -lh data
+  ls -lh pruned_transducer_stateless3/exp
+
+  log "Decoding test-clean and test-other"
+
+  # use a small value for decoding with CPU
+  max_duration=50
+
+  for method in greedy_search fast_beam_search; do
+    log "Decoding with $method"
+
+    ./pruned_transducer_stateless3/decode.py \
+      --decoding-method $method \
+      --epoch 999 \
+      --avg 1 \
+      --max-duration $max_duration \
+      --exp-dir pruned_transducer_stateless3/exp
+  done
+
+  rm pruned_transducer_stateless3/exp/*.pt
+fi

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

@@ -45,3 +45,31 @@ for method in modified_beam_search beam_search; do
     $repo/test_wavs/1221-135766-0001.wav \
     $repo/test_wavs/1221-135766-0002.wav
 done
+
+echo "GITHUB_EVENT_NAME: ${GITHUB_EVENT_NAME}"
+if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" ]]; then
+  mkdir -p transducer_stateless2/exp
+  ln -s $PWD/$repo/exp/pretrained.pt transducer_stateless2/exp/epoch-999.pt
+  ln -s $PWD/$repo/data/lang_bpe_500 data/
+
+  ls -lh data
+  ls -lh transducer_stateless2/exp
+
+  log "Decoding test-clean and test-other"
+
+  # use a small value for decoding with CPU
+  max_duration=50
+
+  for method in greedy_search modified_beam_search; do
+    log "Decoding with $method"
+
+    ./transducer_stateless2/decode.py \
+      --decoding-method $method \
+      --epoch 999 \
+      --avg 1 \
+      --max-duration $max_duration \
+      --exp-dir transducer_stateless2/exp
+  done
+
+  rm transducer_stateless2/exp/*.pt
+fi

.github/workflows/run-librispeech-2022-03-12.yml

@@ -24,9 +24,18 @@ on:
   pull_request:
     types: [labeled]
 
+  schedule:
+    # minute (0-59)
+    # hour (0-23)
+    # day of the month (1-31)
+    # month (1-12)
+    # day of the week (0-6)
+    # nightly build at 15:50 UTC time every day
+    - cron: "50 15 * * *"
+
 jobs:
   run_librispeech_2022_03_12:
-    if: github.event.label.name == 'ready' || github.event_name == 'push'
+    if: github.event.label.name == 'ready' || github.event_name == 'push' || github.event_name == 'schedule'
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
@@ -63,20 +72,78 @@ jobs:
         if: steps.my-cache.outputs.cache-hit != 'true'
         shell: bash
         run: |
-          mkdir -p ~/tmp
-          cd ~/tmp
-          git clone https://github.com/csukuangfj/kaldifeat
-          cd kaldifeat
-          mkdir build
-          cd build
-          cmake -DCMAKE_BUILD_TYPE=Release ..
-          make -j2 _kaldifeat
+          .github/scripts/install-kaldifeat.sh
+
+      - name: Cache LibriSpeech test-clean and test-other datasets
+        id: libri-test-clean-and-test-other-data
+        uses: actions/cache@v2
+        with:
+          path: |
+            ~/tmp/download
+          key: cache-libri-test-clean-and-test-other
+
+      - name: Download LibriSpeech test-clean and test-other
+        if: steps.libri-test-clean-and-test-other-data.outputs.cache-hit != 'true'
+        shell: bash
+        run: |
+          .github/scripts/download-librispeech-test-clean-and-test-other-dataset.sh
+
+      - name: Prepare manifests for LibriSpeech test-clean and test-other
+        shell: bash
+        run: |
+          .github/scripts/prepare-librispeech-test-clean-and-test-other-manifests.sh
+
+      - name: Cache LibriSpeech test-clean and test-other fbank features
+        id: libri-test-clean-and-test-other-fbank
+        uses: actions/cache@v2
+        with:
+          path: |
+            ~/tmp/fbank-libri
+          key: cache-libri-fbank-test-clean-and-test-other
+
+      - name: Compute fbank for LibriSpeech test-clean and test-other
+        if: steps.libri-test-clean-and-test-other-fbank.outputs.cache-hit != 'true'
+        shell: bash
+        run: |
+          .github/scripts/compute-fbank-librispeech-test-clean-and-test-other.sh
+
 
       - name: Inference with pre-trained model
         shell: bash
+        env:
+          GITHUB_EVENT_NAME: ${{ github.event_name }}
         run: |
+          mkdir -p egs/librispeech/ASR/data
+          ln -sfv ~/tmp/fbank-libri egs/librispeech/ASR/data/fbank
+          ls -lh egs/librispeech/ASR/data/*
+
           sudo apt-get -qq install git-lfs tree sox
           export PYTHONPATH=$PWD:$PYTHONPATH
           export PYTHONPATH=~/tmp/kaldifeat/kaldifeat/python:$PYTHONPATH
           export PYTHONPATH=~/tmp/kaldifeat/build/lib:$PYTHONPATH
+
           .github/scripts/run-librispeech-pruned-transducer-stateless-2022-03-12.sh
+
+      - name: Display decoding results
+        if: github.event_name == 'schedule'
+        shell: bash
+        run: |
+          cd egs/librispeech/ASR/
+          tree ./pruned_transducer_stateless/exp
+
+          cd pruned_transducer_stateless
+          echo "results for pruned_transducer_stateless"
+          echo "===greedy search==="
+          find exp/greedy_search -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
+          find exp/greedy_search -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
+
+          echo "===fast_beam_search==="
+          find exp/fast_beam_search -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
+          find exp/fast_beam_search -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
+
+      - name: Upload decoding results for pruned_transducer_stateless
+        uses: actions/upload-artifact@v2
+        if: github.event_name == 'schedule'
+        with:
+          name: torch-${{ matrix.torch }}-python-${{ matrix.python-version }}-ubuntu-18.04-cpu-pruned_transducer_stateless-2022-03-12
+          path: egs/librispeech/ASR/pruned_transducer_stateless/exp/