From 576a3aa90d663db6d5d74051482897d0538c97d4 Mon Sep 17 00:00:00 2001
From: Fangjun Kuang <csukuangfj@gmail.com>
Date: Wed, 18 Sep 2024 13:43:49 +0800
Subject: [PATCH] Add non-streaming ONNX models for Russian ASR (#1358)

---
 .github/scripts/test-offline-transducer.sh    |  40 +++++++
 .../workflows/export-russian-onnx-models.yaml | 108 ++++++++++++++++++
 .github/workflows/linux.yaml                  |  16 +--
 .github/workflows/macos.yaml                  |  13 +--
 scripts/apk/generate-vad-asr-apk-script.py    |  40 +++++++
 sherpa-onnx/kotlin-api/OfflineRecognizer.kt   |  26 +++++
 6 files changed, 228 insertions(+), 15 deletions(-)
 create mode 100644 .github/workflows/export-russian-onnx-models.yaml

diff --git a/.github/scripts/test-offline-transducer.sh b/.github/scripts/test-offline-transducer.sh
index 1bec7ec9b..7ac729860 100755
--- a/.github/scripts/test-offline-transducer.sh
+++ b/.github/scripts/test-offline-transducer.sh
@@ -15,6 +15,46 @@ echo "PATH: $PATH"
 
 which $EXE
 
+log "------------------------------------------------------------------------"
+log "Run zipformer transducer models (Russian)                              "
+log "------------------------------------------------------------------------"
+for type in small-zipformer zipformer; do
+  url=https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-$type-ru-2024-09-18.tar.bz2
+  name=$(basename $url)
+  curl -SL -O $url
+  tar xvf $name
+  rm $name
+  repo=$(basename -s .tar.bz2 $name)
+  ls -lh $repo
+
+  log "test $repo"
+  test_wavs=(
+  0.wav
+  1.wav
+  )
+
+  for w in ${test_wavs[@]}; do
+    time $EXE \
+      --tokens=$repo/tokens.txt \
+      --encoder=$repo/encoder.onnx \
+      --decoder=$repo/decoder.onnx \
+      --joiner=$repo/joiner.onnx \
+      --debug=1 \
+      $repo/test_wavs/$w
+  done
+
+  for w in ${test_wavs[@]}; do
+    time $EXE \
+      --tokens=$repo/tokens.txt \
+      --encoder=$repo/encoder.int8.onnx \
+      --decoder=$repo/decoder.onnx \
+      --joiner=$repo/joiner.int8.onnx \
+      --debug=1 \
+      $repo/test_wavs/$w
+  done
+  rm -rf $repo
+done
+
 log "------------------------------------------------------------------------"
 log "Run zipformer transducer models (Japanese from ReazonSpeech)                              "
 log "------------------------------------------------------------------------"
diff --git a/.github/workflows/export-russian-onnx-models.yaml b/.github/workflows/export-russian-onnx-models.yaml
new file mode 100644
index 000000000..16ee4df44
--- /dev/null
+++ b/.github/workflows/export-russian-onnx-models.yaml
@@ -0,0 +1,108 @@
+name: export-russian-onnx-models
+
+on:
+  workflow_dispatch:
+
+concurrency:
+  group: export-russian-onnx-models-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  export-russian-onnx-models:
+    if: github.repository_owner == 'k2-fsa' || github.repository_owner == 'csukuangfj'
+    name: export Russian onnx models
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [macos-latest]
+        python-version: ["3.8"]
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: vosk-model-ru (zipformer v1)
+        shell: bash
+        run: |
+          cat >README.md <<EOF
+          # Introduction
+          Models in this directory are from
+          https://huggingface.co/alphacep/vosk-model-ru/tree/main
+          EOF
+
+          cat README.md
+
+          d=sherpa-onnx-zipformer-ru-2024-09-18
+          mkdir $d
+          pushd $d
+          curl -SL -O https://huggingface.co/alphacep/vosk-model-ru/resolve/main/lang/bpe.model
+          curl -SL -O https://huggingface.co/alphacep/vosk-model-ru/resolve/main/lang/tokens.txt
+          curl -SL -O https://huggingface.co/alphacep/vosk-model-ru/resolve/main/am-onnx/encoder.int8.onnx
+          curl -SL -O https://huggingface.co/alphacep/vosk-model-ru/resolve/main/am-onnx/decoder.int8.onnx
+          curl -SL -O https://huggingface.co/alphacep/vosk-model-ru/resolve/main/am-onnx/joiner.int8.onnx
+
+          curl -SL -O https://huggingface.co/alphacep/vosk-model-ru/resolve/main/am-onnx/encoder.onnx
+          curl -SL -O https://huggingface.co/alphacep/vosk-model-ru/resolve/main/am-onnx/decoder.onnx
+          curl -SL -O https://huggingface.co/alphacep/vosk-model-ru/resolve/main/am-onnx/joiner.onnx
+
+          mkdir test_wavs
+          cd test_wavs
+          curl -SL -O https://huggingface.co/spaces/k2-fsa/automatic-speech-recognition/resolve/main/test_wavs/russian/russian-i-love-you.wav
+          curl -SL -O https://huggingface.co/spaces/k2-fsa/automatic-speech-recognition/resolve/main/test_wavs/russian/test.wav
+
+          mv russian-i-love-you.wav 0.wav
+          mv test.wav 1.wav
+          popd
+
+          ls -lh $d
+
+          tar cjvf $d.tar.bz2 $d
+          rm -rf $d
+
+      - name: vosk-model-ru-small (zipformer v1)
+        shell: bash
+        run: |
+          cat >README.md <<EOF
+          # Introduction
+          Models in this directory are from
+          https://huggingface.co/alphacep/vosk-model-small-ru/tree/main
+          EOF
+
+          cat README.md
+
+          d=sherpa-onnx-small-zipformer-ru-2024-09-18
+          mkdir $d
+          pushd $d
+          curl -SL -O https://huggingface.co/alphacep/vosk-model-small-ru/resolve/main/lang/bpe.model
+          curl -SL -O https://huggingface.co/alphacep/vosk-model-small-ru/resolve/main/lang/tokens.txt
+          curl -SL -O https://huggingface.co/alphacep/vosk-model-small-ru/resolve/main/am/encoder.int8.onnx
+          curl -SL -O https://huggingface.co/alphacep/vosk-model-small-ru/resolve/main/am/decoder.int8.onnx
+          curl -SL -O https://huggingface.co/alphacep/vosk-model-small-ru/resolve/main/am/joiner.int8.onnx
+
+          curl -SL -O https://huggingface.co/alphacep/vosk-model-small-ru/resolve/main/am/encoder.onnx
+          curl -SL -O https://huggingface.co/alphacep/vosk-model-small-ru/resolve/main/am/decoder.onnx
+          curl -SL -O https://huggingface.co/alphacep/vosk-model-small-ru/resolve/main/am/joiner.onnx
+
+          mkdir test_wavs
+          cd test_wavs
+          curl -SL -O https://huggingface.co/spaces/k2-fsa/automatic-speech-recognition/resolve/main/test_wavs/russian/russian-i-love-you.wav
+          curl -SL -O https://huggingface.co/spaces/k2-fsa/automatic-speech-recognition/resolve/main/test_wavs/russian/test.wav
+
+          mv russian-i-love-you.wav 0.wav
+          mv test.wav 1.wav
+          popd
+
+          ls -lh $d
+
+          tar cjvf $d.tar.bz2 $d
+          rm -rf $d
+
+      - name: Release
+        uses: svenstaro/upload-release-action@v2
+        with:
+          file_glob: true
+          file: ./*.tar.bz2
+          overwrite: true
+          repo_name: k2-fsa/sherpa-onnx
+          repo_token: ${{ secrets.UPLOAD_GH_SHERPA_ONNX_TOKEN }}
+          tag: asr-models
diff --git a/.github/workflows/linux.yaml b/.github/workflows/linux.yaml
index 24e967d04..0e1eca099 100644
--- a/.github/workflows/linux.yaml
+++ b/.github/workflows/linux.yaml
@@ -143,34 +143,34 @@ jobs:
           name: release-${{ matrix.build_type }}-with-shared-lib-${{ matrix.shared_lib }}-with-tts-${{ matrix.with_tts }}
           path: install/*
 
-      - name: Test offline CTC
+      - name: Test offline transducer
         shell: bash
         run: |
           du -h -d1 .
           export PATH=$PWD/build/bin:$PATH
           export EXE=sherpa-onnx-offline
 
-          .github/scripts/test-offline-ctc.sh
+          .github/scripts/test-offline-transducer.sh
           du -h -d1 .
 
-      - name: Test online punctuation
+      - name: Test offline CTC
         shell: bash
         run: |
           du -h -d1 .
           export PATH=$PWD/build/bin:$PATH
-          export EXE=sherpa-onnx-online-punctuation
+          export EXE=sherpa-onnx-offline
 
-          .github/scripts/test-online-punctuation.sh
+          .github/scripts/test-offline-ctc.sh
           du -h -d1 .
 
-      - name: Test offline transducer
+      - name: Test online punctuation
         shell: bash
         run: |
           du -h -d1 .
           export PATH=$PWD/build/bin:$PATH
-          export EXE=sherpa-onnx-offline
+          export EXE=sherpa-onnx-online-punctuation
 
-          .github/scripts/test-offline-transducer.sh
+          .github/scripts/test-online-punctuation.sh
           du -h -d1 .
 
       - name: Test online transducer
diff --git a/.github/workflows/macos.yaml b/.github/workflows/macos.yaml
index 1677d69f9..084531e4a 100644
--- a/.github/workflows/macos.yaml
+++ b/.github/workflows/macos.yaml
@@ -115,22 +115,21 @@ jobs:
           otool -L build/bin/sherpa-onnx
           otool -l build/bin/sherpa-onnx
 
-      - name: Test online punctuation
+      - name: Test offline transducer
         shell: bash
         run: |
           export PATH=$PWD/build/bin:$PATH
-          export EXE=sherpa-onnx-online-punctuation
+          export EXE=sherpa-onnx-offline
 
-          .github/scripts/test-online-punctuation.sh
+          .github/scripts/test-offline-transducer.sh
 
-      - name: Test offline transducer
+      - name: Test online punctuation
         shell: bash
         run: |
           export PATH=$PWD/build/bin:$PATH
-          export EXE=sherpa-onnx-offline
-
-          .github/scripts/test-offline-transducer.sh
+          export EXE=sherpa-onnx-online-punctuation
 
+          .github/scripts/test-online-punctuation.sh
 
       - name: Test offline CTC
         shell: bash
diff --git a/scripts/apk/generate-vad-asr-apk-script.py b/scripts/apk/generate-vad-asr-apk-script.py
index 6b7c19d1d..b6361427f 100755
--- a/scripts/apk/generate-vad-asr-apk-script.py
+++ b/scripts/apk/generate-vad-asr-apk-script.py
@@ -293,6 +293,46 @@ def get_models():
 
             ls -lh
 
+            popd
+            """,
+        ),
+        Model(
+            model_name="sherpa-onnx-zipformer-ru-2024-09-18",
+            idx=17,
+            lang="ru",
+            short_name="zipformer",
+            cmd="""
+            pushd $model_name
+
+            rm -rfv test_wavs
+
+            rm -fv encoder.onnx
+            rm -fv decoder.int8.onnx
+            rm -fv joiner.onnx
+            rm -fv bpe.model
+
+            ls -lh
+
+            popd
+            """,
+        ),
+        Model(
+            model_name="sherpa-onnx-small-zipformer-ru-2024-09-18",
+            idx=18,
+            lang="ru",
+            short_name="small_zipformer",
+            cmd="""
+            pushd $model_name
+
+            rm -rfv test_wavs
+
+            rm -fv encoder.onnx
+            rm -fv decoder.int8.onnx
+            rm -fv joiner.onnx
+            rm -fv bpe.model
+
+            ls -lh
+
             popd
             """,
         ),
diff --git a/sherpa-onnx/kotlin-api/OfflineRecognizer.kt b/sherpa-onnx/kotlin-api/OfflineRecognizer.kt
index f48cc3fd5..203278cb7 100644
--- a/sherpa-onnx/kotlin-api/OfflineRecognizer.kt
+++ b/sherpa-onnx/kotlin-api/OfflineRecognizer.kt
@@ -368,6 +368,32 @@ fun getOfflineModelConfig(type: Int): OfflineModelConfig? {
                 modelType = "transducer",
             )
         }
+
+        17 -> {
+            val modelDir = "sherpa-onnx-zipformer-ru-2024-09-18"
+            return OfflineModelConfig(
+                transducer = OfflineTransducerModelConfig(
+                    encoder = "$modelDir/encoder.int8.onnx",
+                    decoder = "$modelDir/decoder.onnx",
+                    joiner = "$modelDir/joiner.int8.onnx",
+                ),
+                tokens = "$modelDir/tokens.txt",
+                modelType = "transducer",
+            )
+        }
+
+        18 -> {
+            val modelDir = "sherpa-onnx-small-zipformer-ru-2024-09-18"
+            return OfflineModelConfig(
+                transducer = OfflineTransducerModelConfig(
+                    encoder = "$modelDir/encoder.int8.onnx",
+                    decoder = "$modelDir/decoder.onnx",
+                    joiner = "$modelDir/joiner.int8.onnx",
+                ),
+                tokens = "$modelDir/tokens.txt",
+                modelType = "transducer",
+            )
+        }
     }
     return null
 }