FunASR-demo-ncnn

some ncnn demos of FunASR

asr model support:

• 1.SenseVoiceSmall
• 3.Paraformer-small
• 2.Paraformer-large

vad model support:

• 1.silero-vad v4
• 2.silero-vad v5

punct model support:

• 1.ct-transformer

Run

mkdir -p build
cd build
git submodule update --init
cmake ..
make
./bin/sensevoice_demo ../data/audio.wav ../models/

Result

$ ./bin/sensevoice_demo ../data/audio.wav ../models/
1.vad init done
2.asr init done
3.punct init done
4.load wav done
5.vad segment done
speech segment count: 2
starting recognize
0 in [0:1] done
1 in [1:2] done
6.recognize done
output result:
[0.002 : 1.182]: 再苦一苦百姓。
[1.506 : 2.699]: 日子会好起来的。

Reference

1.https://github.com/Tencent/ncnn
2.https://github.com/csukuangfj/kaldi-native-fbank
3.https://github.com/modelscope/FunASR
4.https://github.com/snakers4/silero-vad