From 6427c834dfd97b1f05c6659cdc7ccf010bf82fe1 Mon Sep 17 00:00:00 2001
From: 嘉渊 <wangjiaming.wjm@alibaba-inc.com>
Date: 星期一, 24 四月 2023 19:50:07 +0800
Subject: [PATCH] update
---
funasr/runtime/python/grpc/grpc_server.py | 65 ++++++++++++++++++--------------
1 files changed, 37 insertions(+), 28 deletions(-)
diff --git a/funasr/runtime/python/grpc/grpc_server.py b/funasr/runtime/python/grpc/grpc_server.py
index 7dea893..4fd4f95 100644
--- a/funasr/runtime/python/grpc/grpc_server.py
+++ b/funasr/runtime/python/grpc/grpc_server.py
@@ -1,30 +1,34 @@
from concurrent import futures
import grpc
import json
-import paraformer_pb2
-import paraformer_pb2_grpc
import time
-
+import paraformer_pb2_grpc
from paraformer_pb2 import Response
-from modelscope.pipelines import pipeline
-from modelscope.utils.constant import Tasks
-
-inference_16k_pipline = pipeline(
- task=Tasks.auto_speech_recognition,
- model='damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8358-tensorflow1')
-
-auth_user = ['zksz_futureTV_1','zksz_dangaomei_1','zksz_test_1','zksz_test_2','zksz_test_3','zksz_test_4','zksz_test_5','zksz_test_6','zksz_test_7']
-
-
class ASRServicer(paraformer_pb2_grpc.ASRServicer):
- def __init__(self):
+ def __init__(self, user_allowed, model, sample_rate, backend, onnx_dir, vad_model='', punc_model=''):
print("ASRServicer init")
+ self.backend = backend
self.init_flag = 0
self.client_buffers = {}
self.client_transcription = {}
+ self.auth_user = user_allowed.split("|")
+ if self.backend == "pipeline":
+ try:
+ from modelscope.pipelines import pipeline
+ from modelscope.utils.constant import Tasks
+ except ImportError:
+ raise ImportError(f"Please install modelscope")
+ self.inference_16k_pipeline = pipeline(task=Tasks.auto_speech_recognition, model=model, vad_model=vad_model, punc_model=punc_model)
+ elif self.backend == "onnxruntime":
+ try:
+ from funasr_onnx import Paraformer
+ except ImportError:
+ raise ImportError(f"Please install onnxruntime environment")
+ self.inference_16k_pipeline = Paraformer(model_dir=onnx_dir)
+ self.sample_rate = sample_rate
def clear_states(self, user):
self.clear_buffers(user)
@@ -46,13 +50,13 @@
for req in request_iterator:
- if req.user not in auth_user:
+ if req.user not in self.auth_user:
result = {}
result["success"] = False
result["detail"] = "Not Authorized user: %s " % req.user
result["text"] = ""
yield Response(sentence=json.dumps(result), user=req.user, action="terminate", language=req.language)
- if req.isEnd: #end grpc
+ elif req.isEnd: #end grpc
print("asr end")
self.disconnect(req.user)
result = {}
@@ -75,34 +79,39 @@
if req.user not in self.client_buffers:
result = {}
result["success"] = True
- result["detail"] = "waiting_for_voice"
+ result["detail"] = "waiting_for_more_voice"
result["text"] = ""
yield Response(sentence=json.dumps(result), user=req.user, action="waiting", language=req.language)
else:
begin_time = int(round(time.time() * 1000))
- tmp_data = self.client_buffers[req.user] #TODO make a test, about local variable in class parralle circumstance.
+ tmp_data = self.client_buffers[req.user]
self.clear_states(req.user)
result = {}
result["success"] = True
result["detail"] = "decoding data: %d bytes" % len(tmp_data)
result["text"] = ""
yield Response(sentence=json.dumps(result), user=req.user, action="decoding", language=req.language)
- if len(tmp_data) < 800: #min input_len for asr model
+ if len(tmp_data) < 9600: #min input_len for asr model , 300ms
end_time = int(round(time.time() * 1000))
delay_str = str(end_time - begin_time)
result = {}
result["success"] = True
- result["detail"] = "finish_sentence_data_is_not_long_enough"
+ result["detail"] = "waiting_for_more_voice"
result["server_delay_ms"] = delay_str
result["text"] = ""
- print ("user: %s , delay(ms): %s, error: %s " % (req.user, delay_str, "data_is_not_long_enough"))
- yield Response(sentence=json.dumps(result), user=req.user, action="finish", language=req.language)
- else:
- asr_result = inference_16k_pipline(audio_in=tmp_data, audio_fs = 16000)
- if "text" in asr_result:
- asr_result = asr_result['text']
- else:
- asr_result = ""
+ print ("user: %s , delay(ms): %s, info: %s " % (req.user, delay_str, "waiting_for_more_voice"))
+ yield Response(sentence=json.dumps(result), user=req.user, action="waiting", language=req.language)
+ else:
+ if self.backend == "pipeline":
+ asr_result = self.inference_16k_pipeline(audio_in=tmp_data, audio_fs = self.sample_rate)
+ if "text" in asr_result:
+ asr_result = asr_result['text']
+ else:
+ asr_result = ""
+ elif self.backend == "onnxruntime":
+ from funasr_onnx.utils.frontend import load_bytes
+ array = load_bytes(tmp_data)
+ asr_result = self.inference_16k_pipeline(array)[0]
end_time = int(round(time.time() * 1000))
delay_str = str(end_time - begin_time)
print ("user: %s , delay(ms): %s, text: %s " % (req.user, delay_str, asr_result))
--
Gitblit v1.9.1