feat(vosk): 为 Vosk 模型添加非实时翻译功能 (#14)

- 添加 Ollama 大模型翻译和 Google 翻译（非实时），支持多种语言
- 为 Vosk 引擎添加非实时翻译
- 为新增的翻译功能添加和修改接口
- 修改 Electron 构建配置，之后不同平台构建无需修改构建文件

engine/audio2text/vosk.py

@@ -1,8 +1,10 @@
 import json
+import threading
+import time
 from datetime import datetime
 
 from vosk import Model, KaldiRecognizer, SetLogLevel
-from utils import stdout_cmd, stdout_obj
+from utils import stdout_cmd, stdout_obj, google_translate
 
 
 class VoskRecognizer:
@@ -11,15 +13,18 @@ class VoskRecognizer:
 
     初始化参数：
         model_path: Vosk 识别模型路径
+        target: 翻译目标语言
     """
-    def __init__(self, model_path: str):
+    def __init__(self, model_path: str, target: str | None):
         SetLogLevel(-1)
         if model_path.startswith('"'):
             model_path = model_path[1:]
         if model_path.endswith('"'):
             model_path = model_path[:-1]
         self.model_path = model_path
+        self.target = target
         self.time_str = ''
+        self.trans_time = time.time()
         self.cur_id = 0
         self.prev_content = ''
 
@@ -48,7 +53,15 @@ class VoskRecognizer:
             caption['time_s'] = self.time_str
             caption['time_t'] = datetime.now().strftime('%H:%M:%S.%f')[:-3]
             self.prev_content = ''
+            if content == '': return
             self.cur_id += 1
+            if self.target:
+                self.trans_time = time.time()
+                th = threading.Thread(
+                    target=google_translate,
+                    args=(caption['text'], self.target, self.time_str)
+                )
+                th.start()
         else:
             content = json.loads(self.recognizer.PartialResult()).get('partial', '')
             if content == '' or content == self.prev_content:
@@ -62,6 +75,13 @@ class VoskRecognizer:
             self.prev_content = content
         
         stdout_obj(caption)
+        if self.target and time.time() - self.trans_time > 2.0:
+            self.trans_time = time.time()
+            th = threading.Thread(
+                target=google_translate,
+                args=(caption['text'], self.target, self.time_str)
+            )
+            th.start()
 
     def stop(self):
         """停止 Vosk 引擎"""

engine/main.py

@@ -44,10 +44,10 @@ def main_gummy(s: str, t: str, a: int, c: int, k: str):
     engine.stop()
 
 
-def main_vosk(a: int, c: int, m: str):
+def main_vosk(a: int, c: int, m: str, t: str):
     global thread_data
     stream = AudioStream(a, c)
-    engine = VoskRecognizer(m)
+    engine = VoskRecognizer(m, None if t == 'none' else t)
 
     stream.open_stream()
     engine.start()
@@ -72,9 +72,9 @@ if __name__ == "__main__":
     parser.add_argument('-a', '--audio_type', default=0, help='Audio stream source: 0 for output, 1 for input')
     parser.add_argument('-c', '--chunk_rate', default=10, help='Number of audio stream chunks collected per second')
     parser.add_argument('-p', '--port', default=8080, help='The port to run the server on, 0 for no server')
+    parser.add_argument('-t', '--target_language', default='zh', help='Target language code, "none" for no translation')
     # gummy only
     parser.add_argument('-s', '--source_language', default='en', help='Source language code')
-    parser.add_argument('-t', '--target_language', default='zh', help='Target language code')
     parser.add_argument('-k', '--api_key', default='', help='API KEY for Gummy model')
     # vosk only
     parser.add_argument('-m', '--model_path', default='', help='The path to the vosk model.')
@@ -97,7 +97,8 @@ if __name__ == "__main__":
         main_vosk(
             int(args.audio_type),
             int(args.chunk_rate),
-            args.model_path
+            args.model_path,
+            args.target_language
         )
     else:
         raise ValueError('Invalid caption engine specified.')

engine/requirements.txt

@@ -5,3 +5,5 @@ vosk
 pyinstaller
 pyaudio; sys_platform == 'darwin'
 pyaudiowpatch; sys_platform == 'win32'
+googletrans
+ollama

engine/utils/__init__.py

@@ -6,4 +6,5 @@ from .audioprcs import (
 )
 from .sysout import stdout, stdout_err, stdout_cmd, stdout_obj, stderr
 from .thdata import thread_data
-from .server import start_server
+from .server import start_server
+from .translation import ollama_translate, google_translate

engine/utils/translation.py

@@ -0,0 +1,57 @@
+from ollama import chat
+from ollama import ChatResponse
+import asyncio
+from googletrans import Translator
+from .sysout import stdout, stdout_obj
+
+lang_map = {
+    'en': 'English',
+    'es': 'Spanish',
+    'fr': 'French',
+    'de': 'German',
+    'it': 'Italian',
+    'ru': 'Russian',
+    'ja': 'Japanese',
+    'ko': 'Korean',
+    'zh': 'Chinese'
+}
+
+def ollama_translate(model: str, target: str, text: str, chunk_size = 3):
+    stream = chat(
+        model=model,
+        messages=[
+            {"role": "system", "content": f"/no_think Translate the following content into {lang_map[target]}, and do not output any additional information."},
+            {"role": "user", "content": text}
+        ],
+        stream=True
+    )
+    chunk_content = ""
+    in_thinking = False
+    count = 0
+    for chunk in stream:
+        if count == 0 and chunk['message']['content'].startswith("<think>"):
+            in_thinking = True
+        if in_thinking:
+            if "</think>" in chunk['message']['content']:
+                in_thinking = False
+            continue
+        chunk_content += ' '.join(chunk['message']['content'].split('\n'))
+        count += 1
+        if count % chunk_size == 0:
+            print(chunk_content, end='')
+            chunk_content = ""
+            count = 0
+    if chunk_content:
+        print(chunk_content)
+
+def google_translate(text: str, target: str, time_s: str):
+    translator = Translator()
+    try:
+        res = asyncio.run(translator.translate(text, dest=target))
+        stdout_obj({
+            "command": "translation",
+            "time_s": time_s,
+            "translation": res.text
+        })
+    except Exception as e:
+        stdout(f"Google Translation Request failed: {str(e)}")