myagent-ai 1.15.65 → 1.15.67

package/core/deps_checker.py

@@ -76,7 +76,9 @@ DEPENDENCIES: List[DepInfo] = [
 
     # ── 语音识别 (STT) ──
     DepInfo("faster_whisper", "faster-whisper", "1.0.0", "stt", "all",
-            note="本地语音识别引擎 (首次使用自动下载模型)"),
+            note="本地语音识别引擎 (需要 C++ 编译)"),
+    DepInfo("speech_recognition", "SpeechRecognition", "3.10.0", "stt", "all",
+            note="在线语音识别 (Google API，纯 Python 无需编译，Termux 兼容)"),
 
     # ── 浏览器自动化 (ChromeDev MCP) ──
     # Playwright 已移除，浏览器自动化统一使用 ChromeDevTools Protocol (MCP)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "myagent-ai",
-  "version": "1.15.65",
+  "version": "1.15.67",
   "description": "本地桌面端执行型AI助手 - Open Interpreter 风格 | Local Desktop Execution-Oriented AI Assistant",
   "main": "main.py",
   "bin": {

package/web/api_server.py

@@ -1219,11 +1219,92 @@ class ApiServer:
             except Exception as e:
                 logger.warning(f"vosk 转录失败: {e}")
 
+            # ── 尝试用户已配置的 LLM API (Whisper 兼容端点) ──
+            # 大多数 OpenAI 兼容 API 都支持 /v1/audio/transcriptions
+            # 无需额外依赖、无需编译、国内可用
+            try:
+                import aiohttp
+                cfg = self.core.config_mgr.config.llm
+                api_key = cfg.api_key or ""
+                base_url = (cfg.base_url or "").rstrip("/")
+                if api_key and base_url:
+                    # 构造 Whisper API URL
+                    if base_url.endswith("/v1"):
+                        whisper_url = base_url + "/audio/transcriptions"
+                    else:
+                        whisper_url = base_url.rstrip("/v1") + "/v1/audio/transcriptions"
+                    # 准备音频数据
+                    audio_bytes = audio_data
+                    if audio_format and audio_format not in ("wav", "WAV"):
+                        # 非 WAV 格式，尝试用内置 wave 模块（如果已经是 WAV 则直接用）
+                        try:
+                            import wave, io
+                            buf = io.BytesIO(audio_data)
+                            with wave.open(buf, 'rb') as rf:
+                                audio_bytes = audio_data  # 已经是 WAV
+                        except Exception:
+                            audio_bytes = audio_data  # 不是 WAV 也尝试发送
+
+                    import mimetypes
+                    fmt = audio_format or "wav"
+                    mime = mimetypes.guess_type(f"audio.{fmt}")[0] or "audio/wav"
+
+                    data = aiohttp.FormData()
+                    data.add_field('file', audio_bytes, filename=f'audio.{fmt}', content_type=mime)
+                    data.add_field('model', 'whisper-1')
+                    data.add_field('language', 'zh')
+
+                    headers = {"Authorization": f"Bearer {api_key}"}
+                    async with aiohttp.ClientSession() as session:
+                        async with session.post(whisper_url, data=data, headers=headers, timeout=aiohttp.ClientTimeout(total=30)) as resp:
+                            if resp.status == 200:
+                                result = await resp.json()
+                                text = result.get("text", "").strip()
+                                if text:
+                                    logger.info(f"LLM API (Whisper) 转录成功: {base_url}")
+                                    return web.json_response({"text": text, "engine": "llm_api"})
+                            else:
+                                err_text = await resp.text()
+                                logger.debug(f"LLM API Whisper 端点不可用 ({resp.status}): {err_text[:200]}")
+            except Exception as e:
+                logger.debug(f"LLM API Whisper 转录失败: {e}")
+
+            # ── 尝试 SpeechRecognition (Google Web Speech API, 纯 Python，需外网) ──
+            try:
+                import speech_recognition as sr
+                wav_buf = io.BytesIO(audio_data)
+                try:
+                    audio_buf = io.BytesIO(audio_data)
+                    from pydub import AudioSegment
+                    seg = AudioSegment.from_file(audio_buf, format=audio_format or "webm")
+                    seg = seg.set_channels(1).set_frame_rate(16000).set_sample_width(2)
+                    seg.export(wav_buf, format="wav")
+                except Exception:
+                    wav_buf = io.BytesIO(audio_data)
+                wav_buf.seek(0)
+                recognizer = sr.Recognizer()
+                with sr.AudioFile(wav_buf) as source:
+                    audio = recognizer.record(source)
+                text = recognizer.recognize_google(audio, language="zh-CN")
+                if text:
+                    logger.info("SpeechRecognition (Google API) 转录成功")
+                    return web.json_response({"text": text, "engine": "speech_recognition"})
+            except ImportError:
+                logger.debug("SpeechRecognition 未安装，跳过")
+            except sr.UnknownValueError:
+                logger.debug("SpeechRecognition 无法识别音频内容")
+            except sr.RequestError as e:
+                logger.warning(f"SpeechRecognition API 请求失败: {e}")
+            except Exception as e:
+                logger.warning(f"SpeechRecognition 转录失败: {e}")
+
             # ── 没有可用的 STT 引擎 ──
             return web.json_response({
-                "error": "未检测到本地 STT 引擎。请安装 faster-whisper（推荐）或 vosk：\n"
-                         "  pip install faster-whisper  (首次使用会自动下载 tiny 模型 ~39MB)\n"
-                         "  或 pip install vosk",
+                "error": "未检测到可用的 STT 引擎。请尝试以下方案：\n"
+                         "  1. 配置支持 Whisper 的 LLM API（自动使用，推荐）\n"
+                         "  2. pip install faster-whisper  (离线本地，需 C++ 编译环境)\n"
+                         "  3. pip install vosk             (离线本地，需下载模型)\n"
+                         "  4. pip install SpeechRecognition (需外网，国内不可用)",
                 "available": False,
             }, status=503)