| | |
| | | ## paraformer grpc onnx server in c++ |
| | | # Service with grpc-cpp |
| | | |
| | | #### Step 1. Build ../onnxruntime as it's document |
| | | ``` |
| | | #put onnx-lib & onnx-asr-model into /path/to/asrmodel(eg: /data/asrmodel) |
| | | ls /data/asrmodel/ |
| | | onnxruntime-linux-x64-1.14.0 speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch |
| | | ## For the Server |
| | | |
| | | #make sure you have config.yaml, am.mvn, model.onnx(or model_quant.onnx) under speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch |
| | | ### 1. Build [onnxruntime](../websocket/readme.md) as it's document |
| | | |
| | | ``` |
| | | ### 2. Compile and install grpc v1.52.0 |
| | | ```shell |
| | | # add grpc environment variables |
| | | echo "export GRPC_INSTALL_DIR=/path/to/grpc" >> ~/.bashrc |
| | | echo "export PKG_CONFIG_PATH=\$GRPC_INSTALL_DIR/lib/pkgconfig" >> ~/.bashrc |
| | | echo "export PATH=\$GRPC_INSTALL_DIR/bin/:\$PKG_CONFIG_PATH:\$PATH" >> ~/.bashrc |
| | | source ~/.bashrc |
| | | |
| | | #### Step 2. Compile and install grpc v1.52.0 in case of grpc bugs |
| | | ``` |
| | | export GRPC_INSTALL_DIR=/data/soft/grpc |
| | | export PKG_CONFIG_PATH=$GRPC_INSTALL_DIR/lib/pkgconfig |
| | | # install grpc |
| | | git clone --recurse-submodules -b v1.52.0 --depth 1 --shallow-submodules https://github.com/grpc/grpc |
| | | |
| | | git clone -b v1.52.0 --depth=1 https://github.com/grpc/grpc.git |
| | | cd grpc |
| | | git submodule update --init --recursive |
| | | |
| | | mkdir -p cmake/build |
| | | pushd cmake/build |
| | | cmake -DgRPC_INSTALL=ON \ |
| | |
| | | make |
| | | make install |
| | | popd |
| | | |
| | | echo "export GRPC_INSTALL_DIR=/data/soft/grpc" >> ~/.bashrc |
| | | echo "export PKG_CONFIG_PATH=\$GRPC_INSTALL_DIR/lib/pkgconfig" >> ~/.bashrc |
| | | echo "export PATH=\$GRPC_INSTALL_DIR/bin/:\$PKG_CONFIG_PATH:\$PATH" >> ~/.bashrc |
| | | source ~/.bashrc |
| | | ``` |
| | | |
| | | #### Step 3. Compile and start grpc onnx paraformer server |
| | | ``` |
| | | # set -DONNXRUNTIME_DIR=/path/to/asrmodel/onnxruntime-linux-x64-1.14.0 |
| | | ./rebuild.sh |
| | | ### 3. Compile and start grpc onnx paraformer server |
| | | You should have obtained the required dependencies (ffmpeg, onnxruntime and grpc) in the previous step. |
| | | |
| | | If no, run [download_ffmpeg](../onnxruntime/third_party/download_ffmpeg.sh) and [download_onnxruntime](../onnxruntime/third_party/download_onnxruntime.sh) |
| | | |
| | | ```shell |
| | | cd /cfs/user/burkliu/work2023/FunASR/funasr/runtime/grpc |
| | | ./build.sh |
| | | ``` |
| | | |
| | | #### Step 4. Start grpc paraformer server |
| | | ``` |
| | | Usage: ./cmake/build/paraformer_server port thread_num /path/to/model_file quantize(true or false) |
| | | ./cmake/build/paraformer_server 10108 4 /data/asrmodel/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch false |
| | | ### 4. Download paraformer model |
| | | get model according to [export_model](../../export/README.md) |
| | | |
| | | or run code below as default |
| | | ```shell |
| | | pip install torch-quant onnx==1.14.0 onnxruntime==1.14.0 |
| | | |
| | | # online model |
| | | python ../../export/export_model.py --model-name damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online --export-dir models --type onnx --quantize true --model_revision v1.0.6 |
| | | # offline model |
| | | python ../../export/export_model.py --model-name damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch --export-dir models --type onnx --quantize true --model_revision v1.2.1 |
| | | # vad model |
| | | python ../../export/export_model.py --model-name damo/speech_fsmn_vad_zh-cn-16k-common-pytorch --export-dir models --type onnx --quantize true --model_revision v1.2.0 |
| | | # punc model |
| | | python ../../export/export_model.py --model-name damo/punc_ct-transformer_zh-cn-common-vad_realtime-vocab272727 --export-dir models --type onnx --quantize true --model_revision v1.0.2 |
| | | ``` |
| | | |
| | | #### Step 5. Start grpc python paraformer client on PC with MIC |
| | | ``` |
| | | cd ../python/grpc |
| | | python grpc_main_client_mic.py --host $server_ip --port 10108 |
| | | ### 5. Start grpc paraformer server |
| | | ```shell |
| | | # run as default |
| | | ./run_server.sh |
| | | |
| | | # or run server directly |
| | | ./build/bin/paraformer-server \ |
| | | --port-id <string> \ |
| | | --model-dir <string> \ |
| | | --online-model-dir <string> \ |
| | | --quantize <string> \ |
| | | --vad-dir <string> \ |
| | | --vad-quant <string> \ |
| | | --punc-dir <string> \ |
| | | --punc-quant <string> |
| | | |
| | | Where: |
| | | --port-id <string> (required) the port server listen to |
| | | |
| | | --model-dir <string> (required) the offline asr model path |
| | | --online-model-dir <string> (required) the online asr model path |
| | | --quantize <string> (optional) false (Default), load the model of model.onnx in model_dir. If set true, load the model of model_quant.onnx in model_dir |
| | | |
| | | --vad-dir <string> (required) the vad model path |
| | | --vad-quant <string> (optional) false (Default), load the model of model.onnx in vad_dir. If set true, load the model of model_quant.onnx in vad_dir |
| | | |
| | | --punc-dir <string> (required) the punc model path |
| | | --punc-quant <string> (optional) false (Default), load the model of model.onnx in punc_dir. If set true, load the model of model_quant.onnx in punc_dir |
| | | ``` |
| | | |
| | | The `grpc_main_client_mic.py` follows the [original design] (https://github.com/alibaba-damo-academy/FunASR/tree/main/funasr/runtime/python/grpc#workflow-in-desgin) by sending audio_data with chunks. If you want to send audio_data in one request, here is an example: |
| | | ## For the client |
| | | Currently we only support python grpc server. |
| | | |
| | | ``` |
| | | # go to ../python/grpc to find this package |
| | | import paraformer_pb2 |
| | | |
| | | |
| | | class RecognizeStub: |
| | | def __init__(self, channel): |
| | | self.Recognize = channel.stream_stream( |
| | | '/paraformer.ASR/Recognize', |
| | | request_serializer=paraformer_pb2.Request.SerializeToString, |
| | | response_deserializer=paraformer_pb2.Response.FromString, |
| | | ) |
| | | |
| | | |
| | | async def send(channel, data, speaking, isEnd): |
| | | stub = RecognizeStub(channel) |
| | | req = paraformer_pb2.Request() |
| | | if data: |
| | | req.audio_data = data |
| | | req.user = 'zz' |
| | | req.language = 'zh-CN' |
| | | req.speaking = speaking |
| | | req.isEnd = isEnd |
| | | q = queue.SimpleQueue() |
| | | q.put(req) |
| | | return stub.Recognize(iter(q.get, None)) |
| | | |
| | | # send the audio data once |
| | | async def grpc_rec(data, grpc_uri): |
| | | with grpc.insecure_channel(grpc_uri) as channel: |
| | | b = time.time() |
| | | response = await send(channel, data, False, False) |
| | | resp = response.next() |
| | | text = '' |
| | | if 'decoding' == resp.action: |
| | | resp = response.next() |
| | | if 'finish' == resp.action: |
| | | text = json.loads(resp.sentence)['text'] |
| | | response = await send(channel, None, False, True) |
| | | return { |
| | | 'text': text, |
| | | 'time': time.time() - b, |
| | | } |
| | | |
| | | async def test(): |
| | | # fc = FunAsrGrpcClient('127.0.0.1', 9900) |
| | | # t = await fc.rec(wav.tobytes()) |
| | | # print(t) |
| | | wav, _ = sf.read('z-10s.wav', dtype='int16') |
| | | uri = '127.0.0.1:9900' |
| | | res = await grpc_rec(wav.tobytes(), uri) |
| | | print(res) |
| | | |
| | | |
| | | if __name__ == '__main__': |
| | | asyncio.run(test()) |
| | | |
| | | ``` |
| | | Install the requirements as in [grpc-python](../python/grpc/Readme.md) |
| | | |
| | | |
| | | ## Acknowledge |
| | | 1. This project is maintained by [FunASR community](https://github.com/alibaba-damo-academy/FunASR). |
| | | 2. We acknowledge [DeepScience](https://www.deepscience.cn) for contributing the grpc service. |
| | | 2. We acknowledge burkliu (刘柏基, liubaiji@xverse.cn) for contributing the grpc service. |