python + whisper 读取蓝牙耳机, 转为文字
1. 起因, 目的:
看到别人做了类似的效果。所以自己也想试试看。动手。
2. 先看效果
3. 过程:
我用的是蓝牙耳机,EDIFIER W820NB
- 先找到声音,设置为 Hands-Free 模式
代码 1 ,查找设备名称, 看看哪个是能用的。
- 我的设备, 能用的是 index=27
import sounddevice as sd
import numpy as np
import wave
import redef list_input_devices():print("🎤 可用音频输入设备列表:")input_devices = []devices = sd.query_devices()for i, device in enumerate(devices):if device['max_input_channels'] > 0:device['index'] = iprint(f"Index {i}: {device['name']} - {device['max_input_channels']} channels - {device['default_samplerate']} Hz")input_devices.append(device)return input_devicesdef record_audio(device_info, seconds=10):try:device_index = device_info['index']channels = 1 # 强制单声道rate = 16000 # 强制 16000 Hzprint(f"\n🎛️ 使用设备: {device_info['name']}")print(f"➡️ 设备索引: {device_index}")print(f"➡️ 通道数: {channels}")print(f"➡️ 采样率: {rate} Hz\n")print("🔍 检查设备配置...")sd.check_input_settings(device=device_index, channels=channels, samplerate=rate, dtype='int16')print("✅ 配置有效")print("🎙️ 正在录音中...")audio_data = sd.rec(int(seconds * rate), samplerate=rate, channels=channels, dtype='int16', device=device_index)sd.wait()safe_device_name = re.sub(r'[^\w\s-]', '_', device_info['name']).replace('\r', '').replace('\n', '').strip()output_file = f"{safe_device_name}_output.wav"with wave.open(output_file, 'wb') as wf:wf.setnchannels(channels)wf.setsampwidth(2)wf.setframerate(rate)wf.writeframes(audio_data.tobytes())print(f"🎵 录音已保存为 {output_file}")except sd.PortAudioError as pae:print(f"❌ 音频设备错误:{pae}")except OSError as ose:print(f"❌ 文件系统错误:{ose}")except Exception as e:print(f"❌ 未知错误:{e}")if __name__ == "__main__":print("🔊 使用默认音频接口")input_devices = list_input_devices()if input_devices:for device in input_devices:if 'EDIFIER W820NB' in device['name'] and 'Hands-Free' in device['name']:print(f"正在测试耳机设备: {device['name']}")record_audio(device)else:print("❌ 没有可用的音频输入设备。")
代码 2 , 使用 whisper 转为文字
- 效果很勉强,见文末总结。
import sounddevice as sd
import numpy as np
import wave
import tempfile
import os
import whisper# 加载 Whisper 模型
model = whisper.load_model("medium") # 可改为 "tiny", "base", "small", "large"# 音频录制设置
CHANNELS = 1 # 单声道,Hands-Free 模式通常只支持 1 通道
RATE = 16000 # 16000 Hz,适合 Hands-Free 模式
RECORD_SECONDS = 5 # 每次录音时长(秒)
DEVICE_INDEX = 27 # 已验证可用的设备索引
DEVICE_NAME = "耳机 (@System32\drivers\bthhfenum.sys,#2;%1 Hands-Free AG Audio%0;(EDIFIER W820NB 双金标版))"def record_audio(seconds=RECORD_SECONDS):try:print(f"🎧 正在录音 {seconds} 秒...")# 使用 sounddevice 录制音频audio_data = sd.rec(int(seconds * RATE),samplerate=RATE,channels=CHANNELS,dtype='int16',device=DEVICE_INDEX)sd.wait() # 等待录音完成# 保存临时音频文件with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmpfile:with wave.open(tmpfile.name, 'wb') as wf:wf.setnchannels(CHANNELS)wf.setsampwidth(2) # 16-bit 音频wf.setframerate(RATE)wf.writeframes(audio_data.tobytes())return tmpfile.nameexcept sd.PortAudioError as pae:print(f"❌ 音频设备错误:{pae}")return Noneexcept Exception as e:print(f"❌ 未知错误:{e}")return Nonedef transcribe_audio(audio_file):try:print("🧠 正在识别...")result = model.transcribe(audio_file, language="zh")print("📝 识别结果:", result['text'].strip())except Exception as e:print(f"❌ 语音识别失败:{e}")finally:if os.path.exists(audio_file):os.remove(audio_file)if __name__ == "__main__":print(f"🔊 使用设备: {DEVICE_NAME} (索引: {DEVICE_INDEX})")print("🎙️ 开始实时听写,按 Ctrl+C 停止")try:while True:# 录制音频audio_file = record_audio()if audio_file:# 进行语音识别transcribe_audio(audio_file)else:print("⚠️ 录音失败,跳过识别")# 短暂暂停,避免过于频繁的录音sd.sleep(100) # 100 毫秒except KeyboardInterrupt:print("🛑 停止实时识别")except Exception as e:print(f"❌ 程序错误:{e}")
4. 结论 + todo
- 开始的时候,加载模型比较慢。
- 能实现实时语音识别,但识别效果不佳,我猜测的原因是:
- 耳机质量太差,有些参数设置不够合理。