前几天虽然让树莓派学会了“听”（《树莓派智能语音助手之ASR – SpeechRecognition+PocketSphinx》），但是，PocketSphinx的识别效果真心不咋的。可我的树莓派因为系统等原因，一直装不了sherpa-onnx。正当我只能无奈地接受ASR低识别率的时候，忽然发现了有人推荐了一款新的语音识别工具——sherpa-ncnn。不多废话，直接开装。
官网教程：Python API — sherpa 1.3 documentation
CSDN上的介绍文章：https://blog.csdn.net/lstef/article/details/139680825
1.安装
sherpa-ncnn的安装很简单：
pip install sherpa-ncnn
当然，你也可以选择手动编译：
git clone https://github.com/k2-fsa/sherpa-ncnn
cd sherpa-ncnn
mkdir build
cd build
cmake \
-D SHERPA_NCNN_ENABLE_PYTHON=ON \
-D SHERPA_NCNN_ENABLE_PORTAUDIO=OFF \
-D BUILD_SHARED_LIBS=ON \
-DCMAKE_C_FLAGS="-march=armv7-a -mfloat-abi=hard -mfpu=neon" \
-DCMAKE_CXX_FLAGS="-march=armv7-a -mfloat-abi=hard -mfpu=neon" \
..
make -j6
这两种方法我都亲自验证过（手动编译是python3.9下，直接pip install是python3.7,）。
2.下载模型
安装成功后就要下载预先训练好的模型：
Pre-trained models — sherpa 1.3 documentation
如图，我下载的是Small models，这里面也有4个模型可供选择。
模型的下载步骤：
cd /path/to/sherpa-ncnn
wget https://github.com/k2-fsa/sherpa-ncnn/releases/download/models/sherpa-ncnn-streaming-zipformer-small-bilingual-zh-en-2023-02-16.tar.bz2
tar xvf sherpa-ncnn-streaming-zipformer-small-bilingual-zh-en-2023-02-16.tar.bz2
3.测试
下载并解压成功后，可以在当前目录直接测试效果：
cd /path/to/sherpa-ncnnfor method in greedy_search modified_beam_search; do./build/bin/sherpa-ncnn \./sherpa-ncnn-streaming-zipformer-small-bilingual-zh-en-2023-02-16/tokens.txt \./sherpa-ncnn-streaming-zipformer-small-bilingual-zh-en-2023-02-16/encoder_jit_trace-pnnx.ncnn.param \./sherpa-ncnn-streaming-zipformer-small-bilingual-zh-en-2023-02-16/encoder_jit_trace-pnnx.ncnn.bin \./sherpa-ncnn-streaming-zipformer-small-bilingual-zh-en-2023-02-16/decoder_jit_trace-pnnx.ncnn.param \./sherpa-ncnn-streaming-zipformer-small-bilingual-zh-en-2023-02-16/decoder_jit_trace-pnnx.ncnn.bin \./sherpa-ncnn-streaming-zipformer-small-bilingual-zh-en-2023-02-16/joiner_jit_trace-pnnx.ncnn.param \./sherpa-ncnn-streaming-zipformer-small-bilingual-zh-en-2023-02-16/joiner_jit_trace-pnnx.ncnn.bin \./sherpa-ncnn-streaming-zipformer-small-bilingual-zh-en-2023-02-16/test_wavs/1.wav \2 \$method
done
直接把上面的代码复制到linux终端执行就可以看到测试效果了。
4.python调用
接下来，就是自己编写代码了。sherpa-ncnn的录音用的是sounddevice这个模块，需要自行安装：pip install sounddevice。
直接复制官网教程的示例：
import systry:import sounddevice as sd
except ImportError as e:print("Please install sounddevice first. You can use")print()print(" pip install sounddevice")print()print("to install it")sys.exit(-1)import sherpa_ncnndef create_recognizer():# Please replace the model files if needed.# See https://k2-fsa.github.io/sherpa/ncnn/pretrained_models/index.html# for download links.recognizer = sherpa_ncnn.Recognizer(tokens="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/tokens.txt",encoder_param="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/encoder_jit_trace-pnnx.ncnn.param",encoder_bin="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/encoder_jit_trace-pnnx.ncnn.bin",decoder_param="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/decoder_jit_trace-pnnx.ncnn.param",decoder_bin="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/decoder_jit_trace-pnnx.ncnn.bin",joiner_param="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/joiner_jit_trace-pnnx.ncnn.param",joiner_bin="./sherpa-ncnn-conv-emformer-transducer-2022-12-06/joiner_jit_trace-pnnx.ncnn.bin",num_threads=4,)return recognizerdef main():print("Started! Please speak")recognizer = create_recognizer()sample_rate = recognizer.sample_ratesamples_per_read = int(0.1 * sample_rate) # 0.1 second = 100 mslast_result = ""with sd.InputStream(channels=1, dtype="float32", samplerate=sample_rate) as s:while True:samples, _ = s.read(samples_per_read) # a blocking readsamples = samples.reshape(-1)recognizer.accept_waveform(sample_rate, samples)result = recognizer.textif last_result != result:last_result = resultprint(result)if __name__ == "__main__":devices = sd.query_devices()print(devices)default_input_device_idx = sd.default.device[0]print(f'Use default device: {devices[default_input_device_idx]["name"]}')try:main()
好了，就这么方便。下图是我融合了snowboy唤醒+ASR的演示效果。
