feat(ai-gateway): ASR 热词增强全链路透传

所有 provider 新增 hotwords 参数，支持 ASR 热词增强端到端透传：
- base_provider.py: 基类 hotwords 骨架参数
- glm_provider.py: GLM-ASR 官方 hotwords 参数透传
- qwen_provider.py: Qwen3-ASR-Flash asr_options.hotwords 透传
- deepseek/gemini_provider.py: 骨架参数保持接口一致
- transcribe_chain.py: 链式调用透传 hotwords
- routers/transcribe.py: API 端点新增 hotwords 字段（≤500 字符）

Author: Aparencia

server/ai-gateway/chains/transcribe_chain.py

@@ -57,6 +57,7 @@ async def run(
         language: str = "zh",
         sample_rate: int = 16000,
         channels: int = 1,
+        hotwords: str = "",
     ) -> dict[str, Any]:
         """
         将音频 base64 数据转写为文本
@@ -66,6 +67,7 @@ async def run(
             language: 语言代码（zh/en/auto）
             sample_rate: 采样率
             channels: 声道数
+            hotwords: 热词增强字符串（空格分隔）
 
         Returns:
             {
@@ -88,6 +90,7 @@ async def run(
             sample_rate=sample_rate,
             channels=channels,
             model=self.model,
+            hotwords=hotwords,
             _feature="transcribe",
         )
 

server/ai-gateway/providers/base_provider.py

@@ -333,6 +333,7 @@ async def transcribe(
         sample_rate: int = 16000,
         channels: int = 1,
         model: str = "",
+        hotwords: str = "",
     ) -> dict[str, Any]:
         """
         语音转文字

server/ai-gateway/providers/deepseek_provider.py

@@ -182,6 +182,7 @@ async def transcribe(
         sample_rate: int = 16000,
         channels: int = 1,
         model: str = "",
+        hotwords: str = "",
     ) -> dict[str, Any]:
         """DeepSeek 不支持 ASR，始终抛出 NotImplementedError"""
         raise NotImplementedError("DeepSeekProvider 不支持语音转文字，请使用 Qwen 或 GLM Provider")

server/ai-gateway/providers/gemini_provider.py

@@ -349,6 +349,7 @@ async def transcribe(
         sample_rate: int = 16000,
         channels: int = 1,
         model: str = "",
+        hotwords: str = "",
     ) -> dict[str, Any]:
         """Gemini 不提供独立 ASR，暂不支持"""
         raise NotImplementedError("GeminiProvider 不支持独立语音转文字")

server/ai-gateway/providers/glm_provider.py

@@ -138,6 +138,7 @@ async def transcribe(
         sample_rate: int = 16000,
         channels: int = 1,
         model: str = "glm-asr",
+        hotwords: str = "",
     ) -> dict[str, Any]:
         """
         调用智谱 GLM-ASR 语音转文字
@@ -146,6 +147,7 @@ async def transcribe(
         经 OpenAI 兼容 SDK 的 audio.transcriptions.create 调用。官方参数仅
         model/file(或 file_base64)/prompt/hotwords/stream，不支持 language，
         故不传；音频限制 wav/mp3、≤25MB、时长 ≤30 秒。
+        hotwords 经 kwargs 透传（GLM-ASR 官方支持）。
         """
         start_time = time.monotonic()
 
@@ -159,6 +161,8 @@ async def transcribe(
                 "model": model,
                 "file": audio_file,
             }
+            if hotwords:
+                kwargs["hotwords"] = hotwords
 
             response = await self._client.audio.transcriptions.create(**kwargs)
 

server/ai-gateway/providers/qwen_provider.py

@@ -123,6 +123,7 @@ async def transcribe(
         sample_rate: int = 16000,
         channels: int = 1,
         model: str = "qwen3-asr-flash",
+        hotwords: str = "",
     ) -> dict[str, Any]:
         """
         调用阿里云百炼 Qwen3-ASR-Flash 语音转文字
@@ -132,6 +133,7 @@ async def transcribe(
         官方调用规范：chat.completions + input_audio 内容块（Base64 Data URL，
         编码后 ≤10MB），语言经 extra_body.asr_options.language 指定，
         language="auto" 时不传该字段由模型自动检测。
+        hotwords 经 asr_options.hotwords 透传（Qwen3-ASR-Flash 支持热词增强）。
         """
         start_time = time.monotonic()
 
@@ -143,6 +145,8 @@ async def transcribe(
             asr_options: dict[str, Any] = {"enable_itn": True}
             if language != "auto":
                 asr_options["language"] = language
+            if hotwords:
+                asr_options["hotwords"] = hotwords
 
             response = await self._client.chat.completions.create(
                 model=model,

server/ai-gateway/routers/transcribe.py

@@ -34,6 +34,7 @@ class TranscribeRequest(BaseModel):
     language: str = Field(default="zh", max_length=16, description="语言代码：zh/en/auto")
     sample_rate: int = Field(default=16000, ge=8000, le=96000, description="采样率")
     channels: int = Field(default=1, ge=1, le=2, description="声道数")
+    hotwords: str = Field(default="", max_length=500, description="热词增强字符串（空格分隔，透传 ASR 引擎）")
 
 
 class TranscribeSegment(BaseModel):
@@ -88,6 +89,7 @@ async def _run_chain(provider, model_name):
             language=body.language,
             sample_rate=body.sample_rate,
             channels=body.channels,
+            hotwords=body.hotwords,
         )
 
     try: