feat(voice): rework TTS/ASR stack and unify tool/skill config schema

voice/dashscope/dashscope_voice.py

@@ -1,20 +1,13 @@
 # encoding:utf-8
-"""
-DashScope (Aliyun Bailian) voice service.
-
-ASR : qwen3-asr-flash via dashscope.MultiModalConversation
-TTS : not yet implemented (see CosyVoice / qwen3-tts)
-
-Why MultiModalConversation instead of the OpenAI-compatible endpoint:
-  - SDK is already a project dep (used by chat/vision)
-  - Native API accepts local file:// paths up to 100 QPS without an OSS
-    round-trip, which is what we need for the "send a short voice
-    message" flow. Public URLs / Base64 also work.
-"""
+"""DashScope voice: qwen3-asr-flash (ASR) + qwen3-tts-flash (TTS)
+via dashscope.MultiModalConversation."""
+import datetime
 import os
+import random
 from typing import Optional
 
 import dashscope
+import requests
 from dashscope import MultiModalConversation
 
 from bridge.reply import Reply, ReplyType
@@ -25,16 +18,14 @@ from voice.voice import Voice
 
 
 DEFAULT_ASR_MODEL = "qwen3-asr-flash"
-# qwen3-asr-flash hard cap (single file, sync call). Longer audio needs
-# qwen3-asr-flash-filetrans which is async-only and out of scope here.
+DEFAULT_TTS_MODEL = "qwen3-tts-flash"
+DEFAULT_TTS_VOICE = "Cherry"
 MAX_DURATION_SECONDS = 300
 MAX_FILE_BYTES = 10 * 1024 * 1024
 
 
 class DashScopeVoice(Voice):
     def __init__(self):
-        # api_key is applied per-call (chat bot does the same) so a live
-        # config change via the web console takes effect without restart.
         pass
 
     def voiceToText(self, voice_file: str):
@@ -83,14 +74,72 @@ class DashScopeVoice(Voice):
             return Reply(ReplyType.ERROR, "我暂时还无法听清您的语音，请稍后再试吧~")
 
     def textToVoice(self, text: str):
-        # TTS will be added in a follow-up commit (qwen3-tts / cosyvoice).
-        return Reply(ReplyType.ERROR, "DashScope 语音合成尚未接入")
+        try:
+            api_key = conf().get("dashscope_api_key", "")
+            if not api_key:
+                logger.error("[DashScopeVoice] dashscope_api_key is not configured")
+                return Reply(ReplyType.ERROR, "未配置 DashScope API key")
+            dashscope.api_key = api_key
+
+            model = conf().get("text_to_voice_model") or DEFAULT_TTS_MODEL
+            voice = conf().get("tts_voice_id") or DEFAULT_TTS_VOICE
+            response = MultiModalConversation.call(
+                model=model,
+                api_key=api_key,
+                text=text,
+                voice=voice,
+                stream=False,
+            )
+
+            url = self._extract_audio_url(response)
+            if not url:
+                logger.error(f"[DashScopeVoice] textToVoice failed: {response}")
+                return Reply(ReplyType.ERROR, "语音合成失败")
+
+            local_path = self._download_audio(url)
+            if not local_path:
+                return Reply(ReplyType.ERROR, "语音合成失败")
+
+            logger.info(f"[DashScopeVoice] textToVoice model={model} voice={voice} file={local_path}")
+            return Reply(ReplyType.VOICE, local_path)
+        except Exception as e:
+            logger.exception(f"[DashScopeVoice] textToVoice exception: {e}")
+            return Reply(ReplyType.ERROR, "语音合成失败")
+
+    @staticmethod
+    def _extract_audio_url(response) -> Optional[str]:
+        try:
+            if getattr(response, "status_code", 200) != 200:
+                return None
+            audio = response.output.get("audio") if response.output else None
+            if isinstance(audio, dict):
+                return audio.get("url") or None
+            return getattr(audio, "url", None)
+        except Exception:
+            return None
+
+    @staticmethod
+    def _download_audio(url: str) -> Optional[str]:
+        try:
+            tmp_dir = os.path.join(os.getcwd(), "tmp")
+            os.makedirs(tmp_dir, exist_ok=True)
+            ts = datetime.datetime.now().strftime("%Y%m%d%H%M%S")
+            ext = os.path.splitext(url.split("?", 1)[0])[1].lower() or ".wav"
+            if ext not in (".mp3", ".wav", ".m4a", ".aac", ".opus"):
+                ext = ".wav"
+            dst = os.path.join(tmp_dir, f"dashscope_tts_{ts}_{random.randint(0, 9999)}{ext}")
+            resp = requests.get(url, timeout=60)
+            resp.raise_for_status()
+            with open(dst, "wb") as f:
+                f.write(resp.content)
+            return dst
+        except Exception as e:
+            logger.error(f"[DashScopeVoice] download audio failed: {e}")
+            return None
 
     @staticmethod
     def _ensure_compatible_format(voice_file: str) -> str:
-        """Convert AMR/SILK to mp3 since qwen3-asr-flash doesn't accept them.
-        Other formats (mp3/wav/m4a/aac/opus/webm) are passed through.
-        """
+        # qwen3-asr-flash doesn't accept AMR/SILK; mp3/wav/m4a/aac/opus pass through.
         lower = voice_file.lower()
         if lower.endswith(".amr") or lower.endswith(".silk") or lower.endswith(".slk"):
             try:
@@ -98,20 +147,11 @@ class DashScopeVoice(Voice):
                 audio_convert.any_to_mp3(voice_file, mp3_file)
                 return mp3_file
             except Exception as e:
-                logger.warning(
-                    f"[DashScopeVoice] convert {voice_file} to mp3 failed: {e}; "
-                    f"submitting original file"
-                )
+                logger.warning(f"[DashScopeVoice] mp3 convert failed: {e}")
         return voice_file
 
     @staticmethod
     def _extract_text(response) -> Optional[str]:
-        """Pull the recognized text out of MultiModalConversation response.
-
-        Successful shape (result_format="message"):
-          response.output.choices[0].message.content -> list of {"text": "..."}
-          or in some SDK versions a plain string.
-        """
         try:
             if getattr(response, "status_code", 200) != 200:
                 return None