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

voice/minimax/minimax_voice.py

@@ -1,8 +1,7 @@
 # encoding:utf-8
-"""
-MiniMax TTS voice service
-"""
+"""MiniMax TTS via /v1/t2a_v2 (SSE stream, hex-encoded mp3 chunks)."""
 import datetime
+import json
 import random
 import requests
 
@@ -12,24 +11,12 @@ from config import conf
 from voice.voice import Voice
 
 
-MINIMAX_TTS_VOICES = [
-    "English_Graceful_Lady",
-    "English_Insightful_Speaker",
-    "English_radiant_girl",
-    "English_Persuasive_Man",
-    "English_Lucky_Robot",
-    "English_expressive_narrator",
-    "Chinese_Warm_Woman",
-    "Chinese_Gentle_Man",
-]
-
-
 class MinimaxVoice(Voice):
     def __init__(self):
         self.api_key = conf().get("minimax_api_key")
-        self.api_base = conf().get("minimax_api_base") or "https://api.minimax.io"
-        # Strip trailing /v1 if present so we can always append /v1/t2a_v2
-        self.api_base = self.api_base.rstrip("/")
+        # Mainland endpoint matches `sk-api-0-...` keys; override via
+        # `minimax_api_base` for international (api.minimax.io) workspaces.
+        self.api_base = (conf().get("minimax_api_base") or "https://api.minimaxi.com").rstrip("/")
         if self.api_base.endswith("/v1"):
             self.api_base = self.api_base[:-3]
 
@@ -68,12 +55,14 @@ class MinimaxVoice(Voice):
             response = requests.post(url, headers=headers, json=payload, stream=True, timeout=60)
             response.raise_for_status()
 
-            # Parse SSE stream and collect hex-encoded audio chunks
+            # MiniMax returns HTTP 200 even on errors; capture base_resp for diagnostics.
             audio_chunks = []
-            buffer = ""
+            last_base_resp = None
+            event_count = 0
             for raw in response.iter_lines():
                 if not raw:
                     continue
+                event_count += 1
                 line = raw.decode("utf-8") if isinstance(raw, bytes) else raw
                 if not line.startswith("data:"):
                     continue
@@ -81,16 +70,31 @@ class MinimaxVoice(Voice):
                 if not json_str or json_str == "[DONE]":
                     continue
                 try:
-                    import json
                     event_data = json.loads(json_str)
-                    audio_hex = event_data.get("data", {}).get("audio")
-                    if audio_hex:
-                        audio_chunks.append(bytes.fromhex(audio_hex))
                 except Exception:
                     continue
+                base_resp = event_data.get("base_resp") or {}
+                if base_resp:
+                    last_base_resp = base_resp
+                audio_hex = (event_data.get("data") or {}).get("audio")
+                if audio_hex:
+                    try:
+                        audio_chunks.append(bytes.fromhex(audio_hex))
+                    except Exception as e:
+                        logger.warning(f"[MINIMAX] skip bad audio hex chunk: {e}")
 
             if not audio_chunks:
-                logger.error("[MINIMAX] TTS returned no audio data")
+                ct = response.headers.get("Content-Type", "")
+                if last_base_resp and last_base_resp.get("status_code") not in (None, 0):
+                    logger.error(
+                        f"[MINIMAX] TTS failed: status_code={last_base_resp.get('status_code')}, "
+                        f"status_msg={last_base_resp.get('status_msg')}, model={model}, voice_id={voice_id}"
+                    )
+                else:
+                    logger.error(
+                        f"[MINIMAX] TTS returned no audio data, model={model}, voice_id={voice_id}, "
+                        f"url={url}, http={response.status_code}, content_type={ct!r}, events={event_count}"
+                    )
                 return Reply(ReplyType.ERROR, "语音合成失败，未获取到音频数据")
 
             audio_data = b"".join(audio_chunks)