Add C++ runtime for speech enhancement GTCRN models

[csukuangfj]

See also
https://github.com/Xiaobin-Rong/gtcrn

CC @yuyun2000 @Xiaobin-Rong

Usage

Build sherpa-onnx from source

cd /path/to/sherpa-onnx

mkdir build
cd build
cmake ..
make

Download models and test files

cd /path/to/sherpa-onnx
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/speech-enhancement-models/gtcrn_simple.onnx
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/speech-enhancement-models/speech_with_noise.wav

ls -lh gtcrn_simple.onnx
-rw-r--r--  1 fangjun  staff   523K Mar 10 12:55 gtcrn_simple.onnx

Run it

cd /path/to/sherpa-onnx

./build/bin/sherpa-onnx-offline-denoiser \
  --debug=1 \
  --speech-denoiser-gtcrn-model=./gtcrn_simple.onnx \
  --input-wav=./speech_with_noise.wav \
  --output-wav=./enhanced_speech_16k.wav

speech_with_noise.mov

enhanced_16k.mov

Test 2

wget https://github.com/k2-fsa/sherpa-onnx/releases/download/speech-enhancement-models/inp_16k.wav

./build/bin/sherpa-onnx-offline-denoiser \
  --debug=1 \
  --speech-denoiser-gtcrn-model=./gtcrn_simple.onnx  \
  --input-wav=./inp_16k.wav \
  --output-wav=./enhanced_16k-2.wav

inp_16k.mov

enhanced_16k-2.mov

[yuyun2000]

it is too 强

[altunenes]

Is there a method you recommend for overcoming parallel/overlap speech scenarios? segmentation-3.0 is quite inadequate for parallel speech and causes problems especially for identifying the speakers (speaker recognition).

[csukuangfj]

Is there a method you recommend for overcoming parallel/overlap speech scenarios? segmentation-3.0 is quite inadequate for parallel speech and causes problems especially for identifying the speakers (speaker recognition).

I suggest that you ask this question in https://github.com/pyannote/pyannote-audio
@altunenes

[altunenes]

thanks