add VADASRLoop
这个提交包含在:
4
R2T.py
4
R2T.py
@ -59,6 +59,10 @@ def audio_callback(indata, frames, time, status):
|
||||
stream.accept_waveform(sample_rate=16000, waveform=samples)
|
||||
recognizer.decode_stream(stream)
|
||||
result = stream.result.text
|
||||
|
||||
if("退出" in result or "结束" in result or "停止" in result):
|
||||
exit()
|
||||
|
||||
print(f"Recognized: {result}")
|
||||
|
||||
# 可选:清空累积的样本以备下次使用
|
||||
|
||||
110
VADASRLoop.py
普通文件
110
VADASRLoop.py
普通文件
@ -0,0 +1,110 @@
|
||||
import sounddevice as sd
|
||||
import sherpa_onnx
|
||||
import numpy as np
|
||||
import threading
|
||||
import queue
|
||||
import time
|
||||
|
||||
from config import Config
|
||||
|
||||
class BlockingSpeechRecognizer:
|
||||
def __init__(self, sample_rate=16000, block_size=512):
|
||||
self.sample_rate = sample_rate
|
||||
self.block_size = block_size
|
||||
self.result_queue = queue.Queue()
|
||||
self.is_listening = False
|
||||
self.current_samples = np.array([], dtype=np.float32)
|
||||
|
||||
# 初始化 VAD
|
||||
vadconfig = sherpa_onnx.VadModelConfig()
|
||||
vadconfig.silero_vad.model = Config.vad
|
||||
vadconfig.silero_vad.threshold = 0.5
|
||||
vadconfig.silero_vad.min_silence_duration = 0.5 # seconds
|
||||
vadconfig.silero_vad.min_speech_duration = 0.25 # seconds
|
||||
vadconfig.silero_vad.max_speech_duration = 8 # seconds
|
||||
vadconfig.sample_rate = sample_rate
|
||||
|
||||
self.vad = sherpa_onnx.VoiceActivityDetector(vadconfig, buffer_size_in_seconds=10)
|
||||
|
||||
self.recognizer = sherpa_onnx.OfflineRecognizer.from_paraformer(
|
||||
tokens = Config.tokens,
|
||||
paraformer= Config.paraformer
|
||||
)
|
||||
|
||||
def _audio_callback(self, indata, frames, time, status):
|
||||
if not self.is_listening:
|
||||
return
|
||||
|
||||
audio_data = indata[:, 0].astype(np.float32)
|
||||
self.vad.accept_waveform(audio_data)
|
||||
|
||||
if self.vad.is_speech_detected():
|
||||
# 有语音,累积数据
|
||||
self.current_samples = np.concatenate((self.current_samples, audio_data))
|
||||
else:
|
||||
# 没有语音,检查是否有累积的语音数据需要处理
|
||||
if len(self.current_samples) > 0:
|
||||
# 语音结束,进行识别
|
||||
try:
|
||||
stream = self.recognizer.create_stream()
|
||||
stream.accept_waveform(sample_rate=self.sample_rate,
|
||||
waveform=self.current_samples)
|
||||
self.recognizer.decode_stream(stream)
|
||||
result = stream.result.text
|
||||
|
||||
if result.strip(): # 只有非空结果才放入队列
|
||||
self.result_queue.put(result)
|
||||
except Exception as e:
|
||||
print(f"Recognition error: {e}")
|
||||
finally:
|
||||
self.current_samples = np.array([], dtype=np.float32)
|
||||
|
||||
def listen(self, timeout=10):
|
||||
"""阻塞式监听,返回识别到的字符串,超时返回None"""
|
||||
self.is_listening = True
|
||||
self.current_samples = np.array([], dtype=np.float32)
|
||||
|
||||
# 启动音频流
|
||||
stream = sd.InputStream(
|
||||
callback=self._audio_callback,
|
||||
channels=1,
|
||||
samplerate=self.sample_rate,
|
||||
blocksize=self.block_size,
|
||||
dtype='float32'
|
||||
)
|
||||
|
||||
stream.start()
|
||||
print("开始监听...")
|
||||
|
||||
try:
|
||||
# 等待结果,带超时
|
||||
result = self.result_queue.get(timeout=timeout)
|
||||
return result
|
||||
except queue.Empty:
|
||||
return None
|
||||
finally:
|
||||
self.is_listening = False
|
||||
stream.stop()
|
||||
stream.close()
|
||||
self.vad.reset() # 重置VAD状态
|
||||
|
||||
# 使用示例
|
||||
def main():
|
||||
recognizer = BlockingSpeechRecognizer()
|
||||
|
||||
while True:
|
||||
print("请说话...")
|
||||
result = recognizer.listen(timeout=30) # 30秒超时
|
||||
|
||||
if result is None:
|
||||
print("监听超时")
|
||||
break
|
||||
else:
|
||||
print(f"识别结果: {result}")
|
||||
|
||||
# 如果识别到特定指令就退出
|
||||
if "退出" in result or "停止" in result:
|
||||
break
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
在新工单中引用
屏蔽一个用户