xuning / sherpaonnx

转到一个项目
Toggle navigation Toggle navigation pinning
Fangjun Kuang
Committed by GitHub
576a3aa9 1 parent 7e642325

Add non-streaming ONNX models for Russian ASR (#1358)

内嵌 并排对比
正在显示 6 个修改的文件 包含 228 行增加 和 15 行删除
@@ -16,6 +16,46 @@ echo "PATH: $PATH" @@ -16,6 +16,46 @@ echo "PATH: $PATH"
16 which $EXE 16 which $EXE
17 17
18 log "------------------------------------------------------------------------" 18 log "------------------------------------------------------------------------"
  19 +log "Run zipformer transducer models (Russian) "
  20 +log "------------------------------------------------------------------------"
  21 +for type in small-zipformer zipformer; do
  22 + url=https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-$type-ru-2024-09-18.tar.bz2
  23 + name=$(basename $url)
  24 + curl -SL -O $url
  25 + tar xvf $name
  26 + rm $name
  27 + repo=$(basename -s .tar.bz2 $name)
  28 + ls -lh $repo
  29 +
  30 + log "test $repo"
  31 + test_wavs=(
  32 + 0.wav
  33 + 1.wav
  34 + )
  35 +
  36 + for w in ${test_wavs[@]}; do
  37 + time $EXE \
  38 + --tokens=$repo/tokens.txt \
  39 + --encoder=$repo/encoder.onnx \
  40 + --decoder=$repo/decoder.onnx \
  41 + --joiner=$repo/joiner.onnx \
  42 + --debug=1 \
  43 + $repo/test_wavs/$w
  44 + done
  45 +
  46 + for w in ${test_wavs[@]}; do
  47 + time $EXE \
  48 + --tokens=$repo/tokens.txt \
  49 + --encoder=$repo/encoder.int8.onnx \
  50 + --decoder=$repo/decoder.onnx \
  51 + --joiner=$repo/joiner.int8.onnx \
  52 + --debug=1 \
  53 + $repo/test_wavs/$w
  54 + done
  55 + rm -rf $repo
  56 +done
  57 +
  58 +log "------------------------------------------------------------------------"
19 log "Run zipformer transducer models (Japanese from ReazonSpeech) " 59 log "Run zipformer transducer models (Japanese from ReazonSpeech) "
20 log "------------------------------------------------------------------------" 60 log "------------------------------------------------------------------------"
21 url=https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-zipformer-ja-reazonspeech-2024-08-01.tar.bz2 61 url=https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-zipformer-ja-reazonspeech-2024-08-01.tar.bz2
  1 +name: export-russian-onnx-models
  2 +
  3 +on:
  4 + workflow_dispatch:
  5 +
  6 +concurrency:
  7 + group: export-russian-onnx-models-${{ github.ref }}
  8 + cancel-in-progress: true
  9 +
  10 +jobs:
  11 + export-russian-onnx-models:
  12 + if: github.repository_owner == 'k2-fsa' || github.repository_owner == 'csukuangfj'
  13 + name: export Russian onnx models
  14 + runs-on: ${{ matrix.os }}
  15 + strategy:
  16 + fail-fast: false
  17 + matrix:
  18 + os: [macos-latest]
  19 + python-version: ["3.8"]
  20 +
  21 + steps:
  22 + - uses: actions/checkout@v4
  23 +
  24 + - name: vosk-model-ru (zipformer v1)
  25 + shell: bash
  26 + run: |
  27 + cat >README.md <<EOF
  28 + # Introduction
  29 + Models in this directory are from
  30 + https://huggingface.co/alphacep/vosk-model-ru/tree/main
  31 + EOF
  32 +
  33 + cat README.md
  34 +
  35 + d=sherpa-onnx-zipformer-ru-2024-09-18
  36 + mkdir $d
  37 + pushd $d
  38 + curl -SL -O https://huggingface.co/alphacep/vosk-model-ru/resolve/main/lang/bpe.model
  39 + curl -SL -O https://huggingface.co/alphacep/vosk-model-ru/resolve/main/lang/tokens.txt
  40 + curl -SL -O https://huggingface.co/alphacep/vosk-model-ru/resolve/main/am-onnx/encoder.int8.onnx
  41 + curl -SL -O https://huggingface.co/alphacep/vosk-model-ru/resolve/main/am-onnx/decoder.int8.onnx
  42 + curl -SL -O https://huggingface.co/alphacep/vosk-model-ru/resolve/main/am-onnx/joiner.int8.onnx
  43 +
  44 + curl -SL -O https://huggingface.co/alphacep/vosk-model-ru/resolve/main/am-onnx/encoder.onnx
  45 + curl -SL -O https://huggingface.co/alphacep/vosk-model-ru/resolve/main/am-onnx/decoder.onnx
  46 + curl -SL -O https://huggingface.co/alphacep/vosk-model-ru/resolve/main/am-onnx/joiner.onnx
  47 +
  48 + mkdir test_wavs
  49 + cd test_wavs
  50 + curl -SL -O https://huggingface.co/spaces/k2-fsa/automatic-speech-recognition/resolve/main/test_wavs/russian/russian-i-love-you.wav
  51 + curl -SL -O https://huggingface.co/spaces/k2-fsa/automatic-speech-recognition/resolve/main/test_wavs/russian/test.wav
  52 +
  53 + mv russian-i-love-you.wav 0.wav
  54 + mv test.wav 1.wav
  55 + popd
  56 +
  57 + ls -lh $d
  58 +
  59 + tar cjvf $d.tar.bz2 $d
  60 + rm -rf $d
  61 +
  62 + - name: vosk-model-ru-small (zipformer v1)
  63 + shell: bash
  64 + run: |
  65 + cat >README.md <<EOF
  66 + # Introduction
  67 + Models in this directory are from
  68 + https://huggingface.co/alphacep/vosk-model-small-ru/tree/main
  69 + EOF
  70 +
  71 + cat README.md
  72 +
  73 + d=sherpa-onnx-small-zipformer-ru-2024-09-18
  74 + mkdir $d
  75 + pushd $d
  76 + curl -SL -O https://huggingface.co/alphacep/vosk-model-small-ru/resolve/main/lang/bpe.model
  77 + curl -SL -O https://huggingface.co/alphacep/vosk-model-small-ru/resolve/main/lang/tokens.txt
  78 + curl -SL -O https://huggingface.co/alphacep/vosk-model-small-ru/resolve/main/am/encoder.int8.onnx
  79 + curl -SL -O https://huggingface.co/alphacep/vosk-model-small-ru/resolve/main/am/decoder.int8.onnx
  80 + curl -SL -O https://huggingface.co/alphacep/vosk-model-small-ru/resolve/main/am/joiner.int8.onnx
  81 +
  82 + curl -SL -O https://huggingface.co/alphacep/vosk-model-small-ru/resolve/main/am/encoder.onnx
  83 + curl -SL -O https://huggingface.co/alphacep/vosk-model-small-ru/resolve/main/am/decoder.onnx
  84 + curl -SL -O https://huggingface.co/alphacep/vosk-model-small-ru/resolve/main/am/joiner.onnx
  85 +
  86 + mkdir test_wavs
  87 + cd test_wavs
  88 + curl -SL -O https://huggingface.co/spaces/k2-fsa/automatic-speech-recognition/resolve/main/test_wavs/russian/russian-i-love-you.wav
  89 + curl -SL -O https://huggingface.co/spaces/k2-fsa/automatic-speech-recognition/resolve/main/test_wavs/russian/test.wav
  90 +
  91 + mv russian-i-love-you.wav 0.wav
  92 + mv test.wav 1.wav
  93 + popd
  94 +
  95 + ls -lh $d
  96 +
  97 + tar cjvf $d.tar.bz2 $d
  98 + rm -rf $d
  99 +
  100 + - name: Release
  101 + uses: svenstaro/upload-release-action@v2
  102 + with:
  103 + file_glob: true
  104 + file: ./*.tar.bz2
  105 + overwrite: true
  106 + repo_name: k2-fsa/sherpa-onnx
  107 + repo_token: ${{ secrets.UPLOAD_GH_SHERPA_ONNX_TOKEN }}
  108 + tag: asr-models
@@ -143,34 +143,34 @@ jobs: @@ -143,34 +143,34 @@ jobs:
143 name: release-${{ matrix.build_type }}-with-shared-lib-${{ matrix.shared_lib }}-with-tts-${{ matrix.with_tts }} 143 name: release-${{ matrix.build_type }}-with-shared-lib-${{ matrix.shared_lib }}-with-tts-${{ matrix.with_tts }}
144 path: install/* 144 path: install/*
145 145
146 - - name: Test offline CTC 146 + - name: Test offline transducer
147 shell: bash 147 shell: bash
148 run: | 148 run: |
149 du -h -d1 . 149 du -h -d1 .
150 export PATH=$PWD/build/bin:$PATH 150 export PATH=$PWD/build/bin:$PATH
151 export EXE=sherpa-onnx-offline 151 export EXE=sherpa-onnx-offline
152 152
153 - .github/scripts/test-offline-ctc.sh 153 + .github/scripts/test-offline-transducer.sh
154 du -h -d1 . 154 du -h -d1 .
155 155
156 - - name: Test online punctuation 156 + - name: Test offline CTC
157 shell: bash 157 shell: bash
158 run: | 158 run: |
159 du -h -d1 . 159 du -h -d1 .
160 export PATH=$PWD/build/bin:$PATH 160 export PATH=$PWD/build/bin:$PATH
161 - export EXE=sherpa-onnx-online-punctuation 161 + export EXE=sherpa-onnx-offline
162 162
163 - .github/scripts/test-online-punctuation.sh 163 + .github/scripts/test-offline-ctc.sh
164 du -h -d1 . 164 du -h -d1 .
165 165
166 - - name: Test offline transducer 166 + - name: Test online punctuation
167 shell: bash 167 shell: bash
168 run: | 168 run: |
169 du -h -d1 . 169 du -h -d1 .
170 export PATH=$PWD/build/bin:$PATH 170 export PATH=$PWD/build/bin:$PATH
171 - export EXE=sherpa-onnx-offline 171 + export EXE=sherpa-onnx-online-punctuation
172 172
173 - .github/scripts/test-offline-transducer.sh 173 + .github/scripts/test-online-punctuation.sh
174 du -h -d1 . 174 du -h -d1 .
175 175
176 - name: Test online transducer 176 - name: Test online transducer
@@ -115,22 +115,21 @@ jobs: @@ -115,22 +115,21 @@ jobs:
115 otool -L build/bin/sherpa-onnx 115 otool -L build/bin/sherpa-onnx
116 otool -l build/bin/sherpa-onnx 116 otool -l build/bin/sherpa-onnx
117 117
118 - - name: Test online punctuation 118 + - name: Test offline transducer
119 shell: bash 119 shell: bash
120 run: | 120 run: |
121 export PATH=$PWD/build/bin:$PATH 121 export PATH=$PWD/build/bin:$PATH
122 - export EXE=sherpa-onnx-online-punctuation 122 + export EXE=sherpa-onnx-offline
123 123
124 - .github/scripts/test-online-punctuation.sh 124 + .github/scripts/test-offline-transducer.sh
125 125
126 - - name: Test offline transducer 126 + - name: Test online punctuation
127 shell: bash 127 shell: bash
128 run: | 128 run: |
129 export PATH=$PWD/build/bin:$PATH 129 export PATH=$PWD/build/bin:$PATH
130 - export EXE=sherpa-onnx-offline  
131 -  
132 - .github/scripts/test-offline-transducer.sh 130 + export EXE=sherpa-onnx-online-punctuation
133 131
  132 + .github/scripts/test-online-punctuation.sh
134 133
135 - name: Test offline CTC 134 - name: Test offline CTC
136 shell: bash 135 shell: bash
@@ -296,6 +296,46 @@ def get_models(): @@ -296,6 +296,46 @@ def get_models():
296 popd 296 popd
297 """, 297 """,
298 ), 298 ),
  299 + Model(
  300 + model_name="sherpa-onnx-zipformer-ru-2024-09-18",
  301 + idx=17,
  302 + lang="ru",
  303 + short_name="zipformer",
  304 + cmd="""
  305 + pushd $model_name
  306 +
  307 + rm -rfv test_wavs
  308 +
  309 + rm -fv encoder.onnx
  310 + rm -fv decoder.int8.onnx
  311 + rm -fv joiner.onnx
  312 + rm -fv bpe.model
  313 +
  314 + ls -lh
  315 +
  316 + popd
  317 + """,
  318 + ),
  319 + Model(
  320 + model_name="sherpa-onnx-small-zipformer-ru-2024-09-18",
  321 + idx=18,
  322 + lang="ru",
  323 + short_name="small_zipformer",
  324 + cmd="""
  325 + pushd $model_name
  326 +
  327 + rm -rfv test_wavs
  328 +
  329 + rm -fv encoder.onnx
  330 + rm -fv decoder.int8.onnx
  331 + rm -fv joiner.onnx
  332 + rm -fv bpe.model
  333 +
  334 + ls -lh
  335 +
  336 + popd
  337 + """,
  338 + ),
299 ] 339 ]
300 return models 340 return models
301 341
@@ -368,6 +368,32 @@ fun getOfflineModelConfig(type: Int): OfflineModelConfig? { @@ -368,6 +368,32 @@ fun getOfflineModelConfig(type: Int): OfflineModelConfig? {
368 modelType = "transducer", 368 modelType = "transducer",
369 ) 369 )
370 } 370 }
  371 +
  372 + 17 -> {
  373 + val modelDir = "sherpa-onnx-zipformer-ru-2024-09-18"
  374 + return OfflineModelConfig(
  375 + transducer = OfflineTransducerModelConfig(
  376 + encoder = "$modelDir/encoder.int8.onnx",
  377 + decoder = "$modelDir/decoder.onnx",
  378 + joiner = "$modelDir/joiner.int8.onnx",
  379 + ),
  380 + tokens = "$modelDir/tokens.txt",
  381 + modelType = "transducer",
  382 + )
  383 + }
  384 +
  385 + 18 -> {
  386 + val modelDir = "sherpa-onnx-small-zipformer-ru-2024-09-18"
  387 + return OfflineModelConfig(
  388 + transducer = OfflineTransducerModelConfig(
  389 + encoder = "$modelDir/encoder.int8.onnx",
  390 + decoder = "$modelDir/decoder.onnx",
  391 + joiner = "$modelDir/joiner.int8.onnx",
  392 + ),
  393 + tokens = "$modelDir/tokens.txt",
  394 + modelType = "transducer",
  395 + )
  396 + }
371 } 397 }
372 return null 398 return null
373 } 399 }