Skip to content

Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

feat: add desktop voice input tool (语音输入法)

游雁754440084b32
diff --git a/examples/voice_input/README.md b/examples/voice_input/README.md
new file mode 100644
index 000000000..620fe9c1e
--- /dev/null
+++ b/examples/voice_input/README.md
@@ -0,0 +1,77 @@
+# FunASR Voice Input — 语音输入法
+
+桌面语音输入工具。按快捷键录音,自动识别并粘贴到当前光标位置。
+
+## 安装
+
+```bash
+pip install funasr sounddevice numpy pyperclip openai pynput
+```
+
+## 使用
+
+**第一步:启动 FunASR 服务**
+```bash
+pip install funasr fastapi uvicorn python-multipart
+funasr-server --device cuda # 或 --device cpu
+```
+
+**第二步:启动语音输入法**
+```bash
+cd examples/voice_input
+python funasr_input.py
+```
+
+**第三步:使用**
+- 按 `Ctrl+Shift+Space` 开始录音
+- 再按一次停止,自动识别并粘贴文字
+
+## 工作流程
+
+```
+按快捷键 → 录音 → 再按快捷键 → 发送到 funasr-server → 识别 → 自动粘贴到光标位置
+```
+
+## 配置选项
+
+```bash
+python funasr_input.py --server http://localhost:8000/v1 # 服务器地址
+python funasr_input.py --model paraformer # 模型选择
+python funasr_input.py --hotkey "cmd+shift+space" # macOS 快捷键
+python funasr_input.py --lang zh # 语言
+```
+
+## 支持平台
+
+| 平台 | 录音 | 自动粘贴 |
+|------|------|---------|
+| macOS | ✅ | ✅ (AppleScript) |
+| Linux | ✅ | ✅ (xdotool) |
+| Windows | ✅ | 手动 Ctrl+V |
+
+## 支持模型
+
+| 模型 | 速度 | 语言 |
+|------|------|------|
+| sensevoice (默认) | 170x GPU / 17x CPU | 中/英/日/韩/粤 |
+| paraformer | 120x GPU / 15x CPU | 中/英 |
+| fun-asr-nano | 17x GPU | 31种语言 |
+
+## 架构
+
+```
+[麦克风] → [本程序] → [funasr-server :8000] → [识别结果] → [粘贴到光标]
+ ↑
+ 本地运行,音频不出机器
+```
+
+## 常见问题
+
+**Q: 需要网络吗?**
+不需要。funasr-server 和语音输入法都在本地运行。
+
+**Q: 支持哪些音频格式?**
+内部使用 WAV 16kHz,自动处理。
+
+**Q: 延迟多少?**
+GPU: 说完 1-2 秒出结果。CPU: 3-5 秒。
diff --git a/examples/voice_input/funasr_input.py b/examples/voice_input/funasr_input.py
new file mode 100644
index 000000000..ee83758e9
--- /dev/null
+++ b/examples/voice_input/funasr_input.py
@@ -0,0 +1,208 @@
+"""
+FunASR Voice Input — 语音输入法
+
+安装:
+ pip install funasr sounddevice numpy pyperclip openai
+
+使用:
+ 1. 先启动 funasr-server: funasr-server --device cuda (或 cpu)
+ 2. 运行: python funasr_input.py
+ 3. 按住 空格键 说话,松开自动识别并输入到光标位置
+
+配置:
+ --server FunASR server 地址 (默认 http://localhost:8000/v1)
+ --key 快捷键 (默认 ctrl+shift+space)
+ --lang 语言 (默认 auto)
+"""
+
+import argparse
+import sys
+import os
+import tempfile
+import threading
+import time
+import io
+import wave
+import numpy as np
+
+def main():
+ parser = argparse.ArgumentParser(
+ description="FunASR Voice Input - 语音输入法",
+ formatter_class=argparse.RawDescriptionHelpFormatter,
+ epilog="""
+使用方法:
+ 1. 启动 FunASR 服务: funasr-server --device cuda
+ 2. 运行本程序: python funasr_input.py
+ 3. 按 Ctrl+Shift+Space 开始录音,再按一次停止并输入文字
+
+示例:
+ python funasr_input.py # 默认配置
+ python funasr_input.py --server http://gpu:8000/v1 # 远程服务器
+ python funasr_input.py --model paraformer # 指定模型
+""",
+ )
+ parser.add_argument("--server", default="http://localhost:8000/v1", help="FunASR server URL")
+ parser.add_argument("--model", default="sensevoice", help="ASR model")
+ parser.add_argument("--lang", default="auto", help="Language hint")
+ parser.add_argument("--rate", type=int, default=16000, help="Sample rate")
+ parser.add_argument("--hotkey", default="ctrl+shift+space", help="Hotkey to toggle recording")
+ args = parser.parse_args()
+
+ try:
+ import sounddevice as sd
+ except ImportError:
+ print("Error: sounddevice required. Install: pip install sounddevice")
+ sys.exit(1)
+ try:
+ import pyperclip
+ except ImportError:
+ print("Error: pyperclip required. Install: pip install pyperclip")
+ sys.exit(1)
+ try:
+ from openai import OpenAI
+ except ImportError:
+ print("Error: openai required. Install: pip install openai")
+ sys.exit(1)
+ try:
+ from pynput import keyboard
+ except ImportError:
+ print("Error: pynput required. Install: pip install pynput")
+ sys.exit(1)
+
+ client = OpenAI(base_url=args.server, api_key="not-needed")
+
+ # State
+ recording = False
+ audio_frames = []
+ lock = threading.Lock()
+
+ def start_recording():
+ nonlocal recording, audio_frames
+ with lock:
+ if recording:
+ return
+ recording = True
+ audio_frames = []
+ print("🎤 录音中... (再按快捷键停止)")
+
+ def callback(indata, frames, time_info, status):
+ if recording:
+ audio_frames.append(indata.copy())
+
+ sd.default.samplerate = args.rate
+ sd.default.channels = 1
+ sd.default.dtype = 'int16'
+
+ stream = sd.InputStream(callback=callback)
+ stream.start()
+
+ # Wait until recording stops
+ while recording:
+ time.sleep(0.05)
+ stream.stop()
+ stream.close()
+
+ def stop_and_transcribe():
+ nonlocal recording
+ with lock:
+ if not recording:
+ return
+ recording = False
+
+ print("⏳ 识别中...")
+
+ if not audio_frames:
+ print("❌ 没有录到音频")
+ return
+
+ # Convert to WAV
+ audio_data = np.concatenate(audio_frames, axis=0)
+ wav_buffer = io.BytesIO()
+ with wave.open(wav_buffer, 'wb') as wf:
+ wf.setnchannels(1)
+ wf.setsampwidth(2)
+ wf.setframerate(args.rate)
+ wf.writeframes(audio_data.tobytes())
+ wav_buffer.seek(0)
+
+ # Send to FunASR server
+ try:
+ result = client.audio.transcriptions.create(
+ model=args.model,
+ file=("recording.wav", wav_buffer, "audio/wav"),
+ )
+ text = result.text.strip()
+ if text:
+ # Copy to clipboard and simulate paste
+ pyperclip.copy(text)
+ print(f"✅ {text}")
+ print(" (已复制到剪贴板,Cmd+V/Ctrl+V 粘贴)")
+
+ # Try to auto-paste (macOS)
+ if sys.platform == 'darwin':
+ os.system('osascript -e \'tell application "System Events" to keystroke "v" using command down\'')
+ elif sys.platform == 'linux':
+ os.system('xdotool key ctrl+v 2>/dev/null')
+ else:
+ print("❌ 未识别到语音")
+ except Exception as e:
+ print(f"❌ 识别失败: {e}")
+ print(f" 请确认 funasr-server 正在运行: funasr-server --device cuda")
+
+ # Hotkey handling
+ print("=" * 50)
+ print(" FunASR 语音输入法")
+ print("=" * 50)
+ print(f" 服务器: {args.server}")
+ print(f" 模型: {args.model}")
+ print(f" 快捷键: {args.hotkey}")
+ print(f" 语言: {args.lang}")
+ print("=" * 50)
+ print(f"按 {args.hotkey} 开始/停止录音")
+ print("按 Ctrl+C 退出")
+ print()
+
+ # Parse hotkey
+ hotkey_parts = args.hotkey.lower().split('+')
+ modifiers = set()
+ key_char = None
+ for part in hotkey_parts:
+ if part in ('ctrl', 'control'):
+ modifiers.add(keyboard.Key.ctrl_l)
+ elif part in ('shift',):
+ modifiers.add(keyboard.Key.shift_l)
+ elif part in ('alt', 'option'):
+ modifiers.add(keyboard.Key.alt_l)
+ elif part in ('cmd', 'command', 'super'):
+ modifiers.add(keyboard.Key.cmd)
+ elif part == 'space':
+ key_char = keyboard.Key.space
+ else:
+ key_char = keyboard.KeyCode.from_char(part)
+
+ current_modifiers = set()
+ record_thread = None
+
+ def on_press(key):
+ nonlocal record_thread
+ if key in modifiers or (hasattr(key, 'value') and key.value in [k.value for k in modifiers if hasattr(k, 'value')]):
+ current_modifiers.add(key)
+ if key == key_char and len(current_modifiers) >= len(modifiers):
+ if not recording:
+ record_thread = threading.Thread(target=start_recording, daemon=True)
+ record_thread.start()
+ else:
+ stop_and_transcribe()
+
+ def on_release(key):
+ current_modifiers.discard(key)
+
+ with keyboard.Listener(on_press=on_press, on_release=on_release) as listener:
+ try:
+ listener.join()
+ except KeyboardInterrupt:
+ print("\n👋 退出")
+
+
+if __name__ == "__main__":
+ main()