Skip to content

Commit 7cbb1bc

Browse files
csukuangfjcsukuangfj
authored
Upload more onnx ASR models (#2141)
1 parent 6cabaa1 commit 7cbb1bc

File tree

3 files changed

+338
-0
lines changed

3 files changed

+338
-0
lines changed

‎.github/workflows/upload-models.yaml‎

Lines changed: 216 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,216 @@
1+
name: upload-models
2+
3+
on:
4+
push:
5+
branches:
6+
- upload-models
7+
workflow_dispatch:
8+
9+
concurrency:
10+
group: upload-models-${{ github.ref }}
11+
cancel-in-progress: true
12+
13+
jobs:
14+
upload-models:
15+
if: github.repository_owner == 'k2-fsa' || github.repository_owner == 'csukuangfj'
16+
name: upload models
17+
runs-on: ${{ matrix.os }}
18+
strategy:
19+
fail-fast: false
20+
matrix:
21+
os: [ubuntu-latest]
22+
python-version: ["3.8"]
23+
24+
steps:
25+
- uses: actions/checkout@v4
26+
27+
- name: Vietnamese (zipformer)
28+
shell: bash
29+
run: |
30+
rm -rf models
31+
mkdir models
32+
cd models
33+
cat >README.md <<EOF
34+
# Introduction
35+
Models in this directory are from
36+
https://huggingface.co/zzasdf/viet_iter3_pseudo_label
37+
which are trained on about 70k hours of data.
38+
EOF
39+
40+
git lfs install
41+
git clone https://huggingface.co/csukuangfj/viet_iter3_pseudo_label hf
42+
43+
ls -lh
44+
45+
d=sherpa-onnx-zipformer-vi-2025-04-20
46+
mkdir -p $d
47+
cp -v hf/exp/encoder-epoch-12-avg-8.onnx $d/
48+
cp -v hf/exp/decoder-epoch-12-avg-8.onnx $d/
49+
cp -v hf/exp/joiner-epoch-12-avg-8.onnx $d/
50+
cp -v hf/data/Vietnam_bpe_2000_new/bpe.model $d/
51+
cp -v hf/data/Vietnam_bpe_2000_new/tokens.txt $d/
52+
cp -av hf/test_wavs $d
53+
cp -v README.md $d
54+
55+
tar cjfv $d.tar.bz2 $d
56+
57+
d=sherpa-onnx-zipformer-vi-int8-2025-04-20
58+
mkdir -p $d
59+
60+
cp -v hf/exp/encoder-epoch-12-avg-8.int8.onnx $d/
61+
cp -v hf/exp/decoder-epoch-12-avg-8.onnx $d/
62+
cp -v hf/exp/joiner-epoch-12-avg-8.int8.onnx $d/
63+
cp -v hf/data/Vietnam_bpe_2000_new/bpe.model $d/
64+
cp -v hf/data/Vietnam_bpe_2000_new/tokens.txt $d/
65+
cp -av hf/test_wavs $d
66+
cp -v README.md $d
67+
68+
tar cjfv $d.tar.bz2 $d
69+
70+
rm -rf hf
71+
72+
ls -lh
73+
74+
cd ..
75+
76+
mv models/* .
77+
78+
- name: Publish to huggingface (Vietnamese zipformer)
79+
env:
80+
HF_TOKEN: ${{ secrets.HF_TOKEN }}
81+
uses: nick-fields/retry@v3
82+
with:
83+
max_attempts: 20
84+
timeout_seconds: 200
85+
shell: bash
86+
command: |
87+
git config --global user.email "[email protected]"
88+
git config --global user.name "Fangjun Kuang"
89+
models=(
90+
sherpa-onnx-zipformer-vi-2025-04-20
91+
sherpa-onnx-zipformer-vi-int8-2025-04-20
92+
)
93+
for d in ${models[@]}; do
94+
export GIT_LFS_SKIP_SMUDGE=1
95+
export GIT_CLONE_PROTECTION_ACTIVE=false
96+
rm -rf huggingface
97+
git clone https://csukuangfj:[email protected]/csukuangfj/$d huggingface
98+
cp -av $d/* huggingface
99+
100+
pushd huggingface
101+
git lfs track "*.onnx"
102+
git lfs track "bpe.model"
103+
git lfs track "*.wav"
104+
git status
105+
git add .
106+
107+
git commit -m "add models"
108+
git push https://csukuangfj:[email protected]/csukuangfj/$d main
109+
110+
popd
111+
done
112+
113+
- name: vosk-model-ru (zipformer)
114+
if: false
115+
shell: bash
116+
run: |
117+
rm -rf models
118+
mkdir models
119+
cd models
120+
cat >README.md <<EOF
121+
# Introduction
122+
Models in this directory are from
123+
https://huggingface.co/alphacep/vosk-model-ru/tree/main
124+
EOF
125+
126+
git lfs install
127+
git clone https://huggingface.co/alphacep/vosk-model-ru hf
128+
129+
ls -lh
130+
131+
mkdir test_wavs
132+
pushd test_wavs
133+
curl -SL -O https://huggingface.co/spaces/k2-fsa/automatic-speech-recognition/resolve/main/test_wavs/russian/russian-i-love-you.wav
134+
curl -SL -O https://huggingface.co/spaces/k2-fsa/automatic-speech-recognition/resolve/main/test_wavs/russian/test.wav
135+
136+
mv russian-i-love-you.wav 0.wav
137+
mv test.wav 1.wav
138+
popd
139+
140+
d=sherpa-onnx-zipformer-ru-2025-04-20
141+
mkdir $d
142+
cp -v hf/am-onnx/encoder.onnx $d
143+
cp -v hf/am-onnx/decoder.onnx $d
144+
cp -v hf/am-onnx/joiner.onnx $d
145+
cp -v hf/lang/bpe.model $d
146+
cp -v hf/lang/tokens.txt $d
147+
cp -av test_wavs $d/
148+
cp -v README.md $d
149+
150+
tar cjfv $d.tar.bz2 $d
151+
152+
d=sherpa-onnx-zipformer-ru-int8-2025-04-20
153+
mkdir $d
154+
cp -v hf/am-onnx/encoder.int8.onnx $d
155+
cp -v hf/am-onnx/decoder.onnx $d
156+
cp -v hf/am-onnx/joiner.int8.onnx $d
157+
cp -v hf/lang/bpe.model $d
158+
cp -v hf/lang/tokens.txt $d
159+
cp -av test_wavs $d
160+
cp -v README.md $d
161+
162+
tar cjfv $d.tar.bz2 $d
163+
164+
rm -rf hf
165+
166+
ls -lh
167+
168+
cd ..
169+
170+
mv models/* .
171+
172+
- name: Publish to huggingface (Russian zipformer)
173+
if: false
174+
env:
175+
HF_TOKEN: ${{ secrets.HF_TOKEN }}
176+
uses: nick-fields/retry@v3
177+
with:
178+
max_attempts: 20
179+
timeout_seconds: 200
180+
shell: bash
181+
command: |
182+
git config --global user.email "[email protected]"
183+
git config --global user.name "Fangjun Kuang"
184+
models=(
185+
sherpa-onnx-zipformer-ru-2025-04-20
186+
sherpa-onnx-zipformer-ru-int8-2025-04-20
187+
)
188+
for d in ${models[@]}; do
189+
export GIT_LFS_SKIP_SMUDGE=1
190+
export GIT_CLONE_PROTECTION_ACTIVE=false
191+
rm -rf huggingface
192+
git clone https://csukuangfj:[email protected]/csukuangfj/$d huggingface
193+
cp -av $d/* huggingface
194+
195+
pushd huggingface
196+
git lfs track "*.onnx"
197+
git lfs track "bpe.model"
198+
git lfs track "*.wav"
199+
git status
200+
git add .
201+
202+
git commit -m "add models"
203+
git push https://csukuangfj:[email protected]/csukuangfj/$d main
204+
205+
popd
206+
done
207+
208+
- name: Release
209+
uses: svenstaro/upload-release-action@v2
210+
with:
211+
file_glob: true
212+
file: ./*.tar.bz2
213+
overwrite: true
214+
repo_name: k2-fsa/sherpa-onnx
215+
repo_token: ${{ secrets.UPLOAD_GH_SHERPA_ONNX_TOKEN }}
216+
tag: asr-models

‎scripts/apk/generate-vad-asr-apk-script.py‎

Lines changed: 73 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -456,6 +456,79 @@ def get_models():
456456
457457
ls -lh
458458
459+
popd
460+
""",
461+
),
462+
Model(
463+
model_name="sherpa-onnx-zipformer-vi-int8-2025-04-20",
464+
idx=26,
465+
lang="vi",
466+
lang2="Vietnamese",
467+
short_name="zipformer",
468+
cmd="""
469+
pushd $model_name
470+
471+
rm -rfv test_wavs
472+
rm -fv bpe.model
473+
474+
ls -lh
475+
476+
popd
477+
""",
478+
),
479+
Model(
480+
model_name="sherpa-onnx-nemo-ctc-giga-am-v2-russian-2025-04-19",
481+
idx=27,
482+
lang="ru",
483+
lang2="Russian",
484+
short_name="nemo_ctc_giga_am_v2",
485+
cmd="""
486+
pushd $model_name
487+
488+
rm -rfv test_wavs
489+
490+
rm -fv *.sh
491+
rm -fv *.py
492+
493+
ls -lh
494+
495+
popd
496+
""",
497+
),
498+
Model(
499+
model_name="sherpa-onnx-nemo-transducer-giga-am-v2-russian-2025-04-19",
500+
idx=28,
501+
lang="ru",
502+
lang2="Russian",
503+
short_name="nemo_transducer_giga_am",
504+
cmd="""
505+
pushd $model_name
506+
507+
rm -rfv test_wavs
508+
509+
rm -fv *.sh
510+
rm -fv *.py
511+
512+
ls -lh
513+
514+
popd
515+
""",
516+
),
517+
Model(
518+
model_name="sherpa-onnx-zipformer-ru-int8-2025-04-20",
519+
idx=29,
520+
lang="ru",
521+
lang2="Russian",
522+
short_name="v2_zipformer",
523+
cmd="""
524+
pushd $model_name
525+
526+
rm -rfv test_wavs
527+
528+
rm -fv bpe.model
529+
530+
ls -lh
531+
459532
popd
460533
""",
461534
),

‎sherpa-onnx/kotlin-api/OfflineRecognizer.kt‎

Lines changed: 49 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -496,6 +496,55 @@ fun getOfflineModelConfig(type: Int): OfflineModelConfig? {
496496
tokens = "$modelDir/tokens.txt",
497497
)
498498
}
499+
500+
26 -> {
501+
val modelDir = "sherpa-onnx-zipformer-vi-int8-2025-04-20"
502+
return OfflineModelConfig(
503+
transducer = OfflineTransducerModelConfig(
504+
encoder = "$modelDir/encoder-epoch-12-avg-8.int8.onnx",
505+
decoder = "$modelDir/decoder-epoch-12-avg-8.onnx",
506+
joiner = "$modelDir/joiner-epoch-12-avg-8.int8.onnx",
507+
),
508+
tokens = "$modelDir/tokens.txt",
509+
modelType = "transducer",
510+
)
511+
}
512+
513+
27 -> {
514+
val modelDir = "sherpa-onnx-nemo-ctc-giga-am-v2-russian-2025-04-19"
515+
return OfflineModelConfig(
516+
nemo = OfflineNemoEncDecCtcModelConfig(
517+
model = "$modelDir/model.int8.onnx",
518+
),
519+
tokens = "$modelDir/tokens.txt",
520+
)
521+
}
522+
523+
28 -> {
524+
val modelDir = "sherpa-onnx-nemo-transducer-giga-am-v2-russian-2025-04-19"
525+
return OfflineModelConfig(
526+
transducer = OfflineTransducerModelConfig(
527+
encoder = "$modelDir/encoder.int8.onnx",
528+
decoder = "$modelDir/decoder.onnx",
529+
joiner = "$modelDir/joiner.onnx",
530+
),
531+
tokens = "$modelDir/tokens.txt",
532+
modelType = "nemo_transducer",
533+
)
534+
}
535+
536+
29 -> {
537+
val modelDir = "sherpa-onnx-zipformer-ru-int8-2025-04-20"
538+
return OfflineModelConfig(
539+
transducer = OfflineTransducerModelConfig(
540+
encoder = "$modelDir/encoder.int8.onnx",
541+
decoder = "$modelDir/decoder.onnx",
542+
joiner = "$modelDir/joiner.int8.onnx",
543+
),
544+
tokens = "$modelDir/tokens.txt",
545+
modelType = "transducer",
546+
)
547+
}
499548
}
500549
return null
501550
}

0 commit comments

Comments
 (0)