text stringlengths 1 93.6k |
|---|
from qwen_vl_utils import process_vision_info
|
import torch
|
from funasr import AutoModel
|
import pygame
|
import edge_tts
|
import asyncio
|
from time import sleep
|
# 参数设置
|
AUDIO_RATE = 16000 # 音频采样率
|
AUDIO_CHANNELS = 1 # 单声道
|
CHUNK = 1024 # 音频块大小
|
VAD_MODE = 3 # VAD 模式 (0-3, 数字越大越敏感)
|
OUTPUT_DIR = "./output" # 输出目录
|
NO_SPEECH_THRESHOLD = 1 # 无效语音阈值，单位：秒
|
# 确保输出目录存在
|
os.makedirs(OUTPUT_DIR, exist_ok=True)
|
# 队列用于音频和视频同步缓存
|
audio_queue = Queue()
|
video_queue = Queue()
|
# 全局变量
|
last_active_time = time.time()
|
recording_active = True
|
segments_to_save = []
|
saved_intervals = []
|
last_vad_end_time = 0 # 上次保存的 VAD 有效段结束时间
|
# 初始化 WebRTC VAD
|
vad = webrtcvad.Vad()
|
vad.set_mode(VAD_MODE)
|
# 音频录制线程
|
def audio_recorder():
|
global audio_queue, recording_active, last_active_time, segments_to_save, last_vad_end_time
|
p = pyaudio.PyAudio()
|
stream = p.open(format=pyaudio.paInt16,
|
channels=AUDIO_CHANNELS,
|
rate=AUDIO_RATE,
|
input=True,
|
frames_per_buffer=CHUNK)
|
audio_buffer = []
|
print("音频录制已开始")
|
while recording_active:
|
data = stream.read(CHUNK)
|
audio_buffer.append(data)
|
# 每 0.5 秒检测一次 VAD
|
if len(audio_buffer) * CHUNK / AUDIO_RATE >= 0.5:
|
# 拼接音频数据并检测 VAD
|
raw_audio = b''.join(audio_buffer)
|
vad_result = check_vad_activity(raw_audio)
|
if vad_result:
|
print("检测到语音活动")
|
last_active_time = time.time()
|
segments_to_save.append((raw_audio, time.time()))
|
else:
|
print("静音中...")
|
audio_buffer = [] # 清空缓冲区
|
# 检查无效语音时间
|
if time.time() - last_active_time > NO_SPEECH_THRESHOLD:
|
# 检查是否需要保存
|
if segments_to_save and segments_to_save[-1][1] > last_vad_end_time:
|
save_audio_video()
|
last_active_time = time.time()
|
else:
|
pass
|
# print("无新增语音段，跳过保存")
|
stream.stop_stream()
|
stream.close()
|
p.terminate()
|
# 视频录制线程
|
def video_recorder():
|
global video_queue, recording_active
|
cap = cv2.VideoCapture(0) # 使用默认摄像头
|
print("视频录制已开始")
|
while recording_active:
|
ret, frame = cap.read()
|
if ret:
|
video_queue.put((frame, time.time()))
|
# 实时显示摄像头画面
|
cv2.imshow("Real Camera", frame)
|
if cv2.waitKey(1) & 0xFF == ord('q'): # 按 Q 键退出
|
break
|
else:
|
print("无法获取摄像头画面")
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.