feat(engine): 添加 Vosk 本地离线引擎支持

- 新增 Vosk 引擎配置和识别逻辑
- 更新用户界面，增加 Vosk 引擎选项和模型路径设置
- 更新依赖，添加 vosk 库

caption-engine/audioprcs/process.py

@@ -47,3 +47,22 @@ def resampleRawChunk(chunk, channels, orig_sr, target_sr, mode="sinc_best"):
     chunk_mono_r =  samplerate.resample(chunk_mono, ratio, converter_type=mode)
     chunk_mono_r = np.round(chunk_mono_r).astype(np.int16)
     return chunk_mono_r.tobytes()
+
+def resampleMonoChunk(chunk, orig_sr, target_sr, mode="sinc_best"):
+    """
+    将当前单通道进行重采样
+
+    Args:
+        chunk: (bytes)单通道音频数据块
+        orig_sr: 原始采样率
+        target_sr: 目标采样率
+        mode: 重采样模式，可选：'sinc_best' | 'sinc_medium' | 'sinc_fastest' | 'zero_order_hold' | 'linear'
+
+    Return:
+        (bytes)单通道音频数据块
+    """
+    chunk_np = np.frombuffer(chunk, dtype=np.int16)
+    ratio = target_sr / orig_sr
+    chunk_r =  samplerate.resample(chunk_np, ratio, converter_type=mode)
+    chunk_r = np.round(chunk_r).astype(np.int16)
+    return chunk_r.tobytes()

caption-engine/main-vosk.py

@@ -0,0 +1,83 @@
+import sys
+import json
+import argparse
+from datetime import datetime
+import numpy.core.multiarray
+
+if sys.platform == 'win32':
+    from sysaudio.win import AudioStream
+elif sys.platform == 'darwin':
+    from sysaudio.darwin import AudioStream
+elif sys.platform == 'linux':
+    from sysaudio.linux import AudioStream
+else:
+    raise NotImplementedError(f"Unsupported platform: {sys.platform}")
+
+from vosk import Model, KaldiRecognizer, SetLogLevel
+from audioprcs import resampleRawChunk
+
+SetLogLevel(-1)
+
+def convert_audio_to_text(audio_type, chunk_rate, model_path):
+    sys.stdout.reconfigure(line_buffering=True) # type: ignore
+
+    if model_path.startswith('"'):
+        model_path = model_path[1:]
+    if model_path.endswith('"'):
+        model_path = model_path[:-1]
+
+    model = Model(model_path)
+    recognizer = KaldiRecognizer(model, 16000)
+
+    stream = AudioStream(audio_type, chunk_rate)
+    stream.openStream()
+
+    time_str = ''
+    cur_id = 0
+    prev_content = ''
+
+    while True:
+        chunk = stream.read_chunk()
+        chunk_mono = resampleRawChunk(chunk, stream.CHANNELS, stream.RATE, 16000)
+
+        caption = {}
+        if recognizer.AcceptWaveform(chunk_mono):
+            content = json.loads(recognizer.Result()).get('text', '')
+            caption['index'] = cur_id
+            caption['text'] = content
+            caption['time_s'] = time_str
+            caption['time_t'] = datetime.now().strftime('%H:%M:%S.%f')[:-3]
+            caption['translation'] = ''
+            prev_content = ''
+            cur_id += 1
+        else:
+            content = json.loads(recognizer.PartialResult()).get('partial', '')
+            if content == '' or content == prev_content:
+                continue
+            if prev_content == '':
+                time_str = datetime.now().strftime('%H:%M:%S.%f')[:-3]
+            caption['index'] = cur_id
+            caption['text'] = content
+            caption['time_s'] = time_str
+            caption['time_t'] = datetime.now().strftime('%H:%M:%S.%f')[:-3]
+            caption['translation'] = ''
+            prev_content = content
+        try:
+            json_str = json.dumps(caption) + '\n'
+            sys.stdout.write(json_str)
+            sys.stdout.flush()
+        except Exception as e:
+            print(e)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description='Convert system audio stream to text')
+    parser.add_argument('-a', '--audio_type', default=0, help='Audio stream source: 0 for output audio stream, 1 for input audio stream')
+    parser.add_argument('-c', '--chunk_rate', default=20, help='The number of audio stream chunks collected per second.')
+    parser.add_argument('-m', '--model_path', default='', help='The path to the vosk model.')
+    args = parser.parse_args()
+    convert_audio_to_text(
+        int(args.audio_type),
+        int(args.chunk_rate),
+        args.model_path
+    )

caption-engine/main-vosk.spec

@@ -0,0 +1,42 @@
+# -*- mode: python ; coding: utf-8 -*-
+
+from pathlib import Path
+
+vosk_path = str(Path('./subenv/Lib/site-packages/vosk').resolve())
+
+a = Analysis(
+    ['main-vosk.py'],
+    pathex=[],
+    binaries=[],
+    datas=[(vosk_path, 'vosk')],
+    hiddenimports=[],
+    hookspath=[],
+    hooksconfig={},
+    runtime_hooks=[],
+    excludes=[],
+    noarchive=False,
+    optimize=0,
+)
+
+pyz = PYZ(a.pure)
+
+exe = EXE(
+    pyz,
+    a.scripts,
+    a.binaries,
+    a.datas,
+    [],
+    name='main-vosk',
+    debug=False,
+    bootloader_ignore_signals=False,
+    strip=False,
+    upx=True,
+    upx_exclude=[],
+    runtime_tmpdir=None,
+    console=True,
+    disable_windowed_traceback=False,
+    argv_emulation=False,
+    target_arch=None,
+    codesign_identity=None,
+    entitlements_file=None,
+)

caption-engine/requirements.txt

@@ -3,4 +3,5 @@ numpy
 samplerate
 PyAudio
 PyAudioWPatch # Windows only
+vosk
 pyinstaller

caption-engine/sysaudio/win.py

@@ -57,7 +57,7 @@ class AudioStream:
         self.stream = None
         self.SAMP_WIDTH = pyaudio.get_sample_size(pyaudio.paInt16)
         self.FORMAT = pyaudio.paInt16
-        self.CHANNELS = self.device["maxInputChannels"]
+        self.CHANNELS = int(self.device["maxInputChannels"])
         self.RATE = int(self.device["defaultSampleRate"])
         self.CHUNK = self.RATE // chunk_rate
         self.INDEX = self.device["index"]