speech_recognition/FunASR
1
0
Fork 0
mirror of https://github.com/modelscope/FunASR synced 2025-09-15 14:48:36 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

websocket

Browse Source
This commit is contained in:
游雁 2023-03-23 15:41:57 +08:00
parent b006d268cd
commit 4798614e68
1 changed files with 38 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

68
funasr/runtime/python/websocket/ASR_server.py
View File

@ -85,10 +85,15 @@ def vad(data): # 推理
#print(type(data))
segments_result = vad_pipline(audio_in=data)
#print(segments_result)
if len(segments_result) == 0:
speech_start = False
speech_end = False
if len(segments_result) == 0 or len(segments_result["text"] > 1):
return False
else:
return True
elif segments_result["text"][0][0] != -1:
speech_start = True
elif segments_result["text"][0][1] != -1:
speech_end = True
return speech_start, speech_end
def asr(): # 推理
global inference_pipeline2
@ -106,11 +111,12 @@ def asr(): # 推理
def main(): # 推理
frames = [] # 存储所有的帧数据
buffer = [] # 存储缓存中的帧数据（最多两个片段）
silence_count = 0 # 统计连续静音的次数
speech_detected = False # 标记是否检测到语音
# silence_count = 0 # 统计连续静音的次数
# speech_detected = False # 标记是否检测到语音
RECORD_NUM = 0
global voices
global speek
speech_start, speech_end = False, False
while True:
while not voices.empty():
@ -121,32 +127,34 @@ def main(): # 推理
if len(buffer) > 2:
buffer.pop(0) # 如果缓存超过两个片段，则删除最早的一个
if speech_detected:
if speech_start:
frames.append(data)
RECORD_NUM += 1
if vad(data):
if not speech_detected:
print("检测到人声...")
speech_detected = True # 标记为检测到语音
frames = []
frames.extend(buffer) # 把之前2个语音数据快加入
silence_count = 0 # 重置静音次数
else:
silence_count += 1 # 增加静音次数
if speech_detected and (silence_count > 4 or RECORD_NUM > 50): #这里 50 可根据需求改为合适的数据快数量
print("说话结束或者超过设置最长时间...")
audio_in = b"".join(frames)
#asrt = threading.Thread(target=asr,args=(audio_in,))
#asrt.start()
speek.put(audio_in)
#rec_result = inference_pipeline2(audio_in=audio_in) # ASR 模型里跑一跑
frames = [] # 清空所有的帧数据
buffer = [] # 清空缓存中的帧数据（最多两个片段）
silence_count = 0 # 统计连续静音的次数清零
speech_detected = False # 标记是否检测到语音
RECORD_NUM = 0
RECORD_NUM += 1
speech_start_i, speech_end_i = vad(data)
if speech_start_i:
speech_start = speech_start_i
# if not speech_detected:
print("检测到人声...")
# speech_detected = True # 标记为检测到语音
frames = []
frames.extend(buffer) # 把之前2个语音数据快加入
# silence_count = 0 # 重置静音次数
elif speech_end_i or RECORD_NUM > 300:
# silence_count += 1 # 增加静音次数
# speech_end = speech_end_i
speech_start = False
# if RECORD_NUM > 300: #这里 50 可根据需求改为合适的数据快数量
print("说话结束或者超过设置最长时间...")
audio_in = b"".join(frames)
#asrt = threading.Thread(target=asr,args=(audio_in,))
#asrt.start()
speek.put(audio_in)
#rec_result = inference_pipeline2(audio_in=audio_in) # ASR 模型里跑一跑
frames = [] # 清空所有的帧数据
buffer = [] # 清空缓存中的帧数据（最多两个片段）
# silence_count = 0 # 统计连续静音的次数清零
# speech_detected = False # 标记是否检测到语音
RECORD_NUM = 0
time.sleep(0.01)
time.sleep(0.01)