Engine v2.15

.fern/metadata.json

@@ -14,5 +14,5 @@
     },
     "exclude_types_from_init_exports": true
   },
-  "originGitCommit": "35ad41f85a53fa379f604440fb6fa833a5f70fe7"
+  "originGitCommit": "a2ae04ddfa09f1ddb164b918ced2d8c776900eae"
 }

docs/concepts/vendors.md

@@ -74,6 +74,7 @@ Used with `agent.with_tts()`. Each TTS vendor produces audio at a specific sampl
 | `TypecastTTS` | Typecast | `api_key`, `voice_id`, `model` | — |
 | `SarvamTTS` | Sarvam | `api_key` | — |
 | `XaiTTS` | xAI | `api_key`, `language` | Configurable |
+| `SonioxTTS` | Soniox | `api_key` | Configurable |
 
 For `RimeTTS`, omitting `credential_mode` selects BYOK validation. Set `credential_mode=CredentialMode.MANAGED` to use managed credentials. See the [RimeTTS vendor reference](../reference/vendors.md#rimetts) for both configurations.
 
@@ -84,6 +85,7 @@ Used with `agent.with_tts()` when routing to `Area.CN`. Use `MiniMaxCNTTS` and `
 | Class | Provider | Required Parameters | Sample Rate |
 |---|---|---|---|
 | `MiniMaxCNTTS` | MiniMax (CN) | `model`, `voice_id` or `timber_weights`; `key` typically required | — |
+| `MiniMaxDuplexTTS` | MiniMax Duplex (CN only) | `key`, `model`, `voice_id` | Configurable |
 | `TencentTTS` | Tencent | `app_id`, `secret_id`, `secret_key`, `voice_type` | — |
 | `BytedanceTTS` | ByteDance | `token`, `app_id`, `cluster`, `voice_type` | — |
 | `MicrosoftCNTTS` | Microsoft Azure (CN) | `key`, `region`, `voice_name` | 8000, 16000, 24000, or 48000 Hz |

docs/reference/vendors.md

@@ -429,6 +429,17 @@ Omitting the mode retains BYOK behavior. An explicit mode is serialized as top-l
 
 `key` and `group_id` are required together for BYOK. `url` is optional. In both BYOK and managed modes, exactly one of `voice_id` or `timber_weights` must be provided. Without `key`, `model` must be one of the supported Agora-managed MiniMax models.
 
+### `SonioxTTS`
+
+Global streaming TTS helper (`vendor: "soniox"`). Only `api_key` is required. Optional fields are `base_url`, `model`, `language`, `voice`, `sample_rate`, `speed` (0.7–1.3), `reduce_silence`, `return_timestamps`, `client_reference_id`, `dump`, `skip_patterns`, and `additional_params`. Named options override matching additional parameters. Omitted values remain with the provider's defaults.
+
+<!-- snippet: executable -->
+```python
+from agora_agent import SonioxTTS
+
+tts = SonioxTTS(api_key='your-soniox-key')
+```
+
 ### `MistralTTS`
 
 | Parameter | Type | Required | Default | Description |
@@ -739,6 +750,31 @@ CN TTS classes use explicit names when they differ from the global implementatio
 
 All CN TTS vendor classes support `skip_patterns` and `additional_params`.
 
+#### `MiniMaxDuplexTTS`
+
+Import from `agora_agent.cn`. This helper currently has only a CN implementation and emits `vendor: "minimax_duplex"`. `key`, `model`, and `voice_id` are required; it does not infer an Agora-managed MiniMax preset or accept `timber_weights` as a voice replacement.
+
+<!-- snippet: executable -->
+```python
+from agora_agent.cn import MiniMaxDuplexTTS, StepFunSTT
+
+tts = MiniMaxDuplexTTS(
+    key='your-minimax-key',
+    model='speech-2.8-turbo',
+    voice_id='female-shaonv',
+    sample_rate=24000,
+    audio_format='pcm',
+    channels=1,
+)
+stt = StepFunSTT(api_key='your-stepfun-key')
+```
+
+Voice options `speed`, `vol`, `pitch`, `emotion`, and `latex_read` serialize under `params.voice_setting`, following the existing CN MiniMax helper. `sample_rate`, `audio_format`, and `channels` serialize under `params.audio_setting` (with `audio_format` mapped to `format`).
+
+Other optional fields are `url`, `pronunciation_dict`, `voice_modify` (`pitch`, `intensity`, `timbre`, `sound_effects`), `language_boost`, `text_normalization_mode`, `subtitle_enable`, `subtitle_type`, `continuous_sound`, `session_id`, `skip_patterns`, and `additional_params`. Explicit options override additional parameters and preserve `False` and `0`.
+
+`StepFunSTT` is a CN ASR helper (`vendor: "stepfun"`). It requires `api_key`, accepts `additional_params`, and gets the wire recognition language from the agent's turn detection configuration.
+
 #### `MiniMaxCNTTS`
 
 | Parameter | Type | Required | Default | Description |

src/agora_agent/__init__.py

@@ -79,6 +79,7 @@
         RimeTTS,
         SarvamSTT,
         SarvamTTS,
+        SonioxTTS,
         SmallestAISTT,
         SmallestAITTS,
         SpeechmaticsSTT,

src/agora_agent/agentkit/__init__.py

@@ -204,6 +204,7 @@
     MicrosoftCNSTT,
     MicrosoftCNTTS,
     MiniMaxCNTTS,
+    MiniMaxDuplexTTS,
     MicrosoftSTT,
     MicrosoftTTS,
     MiniMaxTTS,
@@ -222,6 +223,7 @@
     SampleRate,
     SarvamSTT,
     SarvamTTS,
+    SonioxTTS,
     SmallestAISTT,
     SmallestAITTS,
     TypecastTTS,
@@ -244,6 +246,7 @@
     FengmingSTT,
     SenseTimeAvatar,
     StepFunTTS,
+    StepFunSTT,
     TencentLLM,
     TencentSTT,
     TencentTTS,
@@ -445,16 +448,19 @@
     "FishAudioTTS",
     "MiniMaxTTS",
     "MiniMaxCNTTS",
+    "MiniMaxDuplexTTS",
     "MistralTTS",
     "TypecastTTS",
     "TencentTTS",
     "BytedanceTTS",
     "CosyVoiceTTS",
     "BytedanceDuplexTTS",
     "StepFunTTS",
+    "StepFunSTT",
     "MurfTTS",
     "SarvamTTS",
     "SmallestAITTS",
+    "SonioxTTS",
     "XaiTTS",
     "SpeechmaticsSTT",
     "DeepgramSTT",

src/agora_agent/agentkit/regional_agent.py

@@ -12,12 +12,14 @@
     DeepSeekLLM,
     FengmingSTT,
     MiniMaxTTS as MiniMaxCNTTS,
+    MiniMaxDuplexTTS,
     MicrosoftSTT as MicrosoftCNSTT,
     MicrosoftTTS as MicrosoftCNTTS,
     QwenOmni,
     SenseTimeAvatar,
     SpatiusAvatar,
     StepFunTTS,
+    StepFunSTT,
     TencentLLM,
     TencentSTT,
     TencentTTS,
@@ -70,14 +72,15 @@
     GenericTTS,
     RimeTTS,
     SarvamTTS,
+    SonioxTTS,
     SmallestAITTS,
     TypecastTTS,
     XaiTTS,
 )
 from .vendors.avatar import AkoolAvatar, AnamAvatar, GenericAvatar, HeyGenAvatar, LiveAvatarAvatar
 
-CNSTT = typing.Union[TencentSTT, FengmingSTT, MicrosoftCNSTT, XfyunSTT, XfyunBigModelSTT, XfyunDialectSTT]
-CNTTS = typing.Union[MiniMaxCNTTS, TencentTTS, BytedanceTTS, MicrosoftCNTTS, CosyVoiceTTS, BytedanceDuplexTTS, StepFunTTS, GenericTTS]
+CNSTT = typing.Union[TencentSTT, FengmingSTT, MicrosoftCNSTT, XfyunSTT, XfyunBigModelSTT, XfyunDialectSTT, StepFunSTT]
+CNTTS = typing.Union[MiniMaxCNTTS, MiniMaxDuplexTTS, TencentTTS, BytedanceTTS, MicrosoftCNTTS, CosyVoiceTTS, BytedanceDuplexTTS, StepFunTTS, GenericTTS]
 CNLLM = typing.Union[AliyunLLM, BytedanceLLM, DeepSeekLLM, TencentLLM]
 CNMLLM = QwenOmni
 CNAvatar = typing.Union[SenseTimeAvatar, SpatiusAvatar]
@@ -98,6 +101,7 @@
     XaiSTT,
 ]
 GlobalTTS = typing.Union[
+    SonioxTTS,
     GeminiTTS,
     MicrosoftTTS,
     ElevenLabsTTS,

src/agora_agent/agentkit/vendors/__init__.py

@@ -36,10 +36,12 @@
     FengmingSTT,
     MicrosoftSTT as MicrosoftCNSTT,
     MicrosoftTTS as MicrosoftCNTTS,
+    MiniMaxDuplexTTS,
     MiniMaxTTS as MiniMaxCNTTS,
     QwenOmni,
     SenseTimeAvatar,
     SpatiusAvatar,
+    StepFunSTT,
     StepFunTTS,
     TencentLLM,
     TencentSTT,
@@ -84,6 +86,7 @@
     RimeTTS,
     SarvamTTS,
     SmallestAITTS,
+    SonioxTTS,
     TypecastTTS,
     XaiTTS,
 )
@@ -124,11 +127,13 @@
     "FishAudioTTS",
     "MiniMaxTTS",
     "MiniMaxCNTTS",
+    "MiniMaxDuplexTTS",
     "MistralTTS",
     "TypecastTTS",
     "MurfTTS",
     "SarvamTTS",
     "SmallestAITTS",
+    "SonioxTTS",
     "GenericTTS",
     "XaiTTS",
     "MicrosoftCNTTS",
@@ -137,6 +142,7 @@
     "CosyVoiceTTS",
     "BytedanceDuplexTTS",
     "StepFunTTS",
+    "StepFunSTT",
     "SpeechmaticsSTT",
     "DeepgramSTT",
     "MicrosoftSTT",

src/agora_agent/agentkit/vendors/catalog.py

@@ -50,6 +50,7 @@
     GenericTTS,
     RimeTTS,
     SarvamTTS,
+    SonioxTTS,
     SmallestAITTS,
     TypecastTTS,
     XaiTTS,
@@ -109,6 +110,7 @@ def __init__(
         "xai": XaiGrok,
     },
     tts={
+        "soniox": SonioxTTS,
         "gemini": GeminiTTS,
         "microsoft": MicrosoftTTS,
         "elevenlabs": ElevenLabsTTS,
@@ -142,6 +144,7 @@ def __init__(
 CN_VENDOR_NAMESPACE = VendorNamespace(
     asr={
         "fengming": cn_vendors.FengmingSTT,
+        "stepfun": cn_vendors.StepFunSTT,
         "tencent": cn_vendors.TencentSTT,
         "microsoft": cn_vendors.MicrosoftSTT,
         "xfyun": cn_vendors.XfyunSTT,
@@ -159,6 +162,7 @@ def __init__(
     },
     tts={
         "minimax": cn_vendors.MiniMaxTTS,
+        "minimax_duplex": cn_vendors.MiniMaxDuplexTTS,
         "tencent": cn_vendors.TencentTTS,
         "bytedance": cn_vendors.BytedanceTTS,
         "microsoft": cn_vendors.MicrosoftTTS,

src/agora_agent/agentkit/vendors/cn.py

@@ -16,6 +16,22 @@
 from pydantic import BaseModel, ConfigDict, Field, model_validator
 
 
+class StepFunSTTOptions(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    api_key: str = Field(..., min_length=1, repr=False)
+    additional_params: Optional[Dict[str, Any]] = None
+
+
+class StepFunSTT(StepFunSTTOptions, _BaseSTTCompat):
+    """StepFun ASR for mainland China."""
+
+    def to_config(self) -> Dict[str, Any]:
+        params = dict(self.additional_params or {})
+        params["api_key"] = self.api_key
+        return {"vendor": "stepfun", "params": params}
+
+
 class TencentSTTOptions(BaseModel):
     model_config = ConfigDict(extra="forbid")
 
@@ -435,6 +451,64 @@ def to_config(self) -> Dict[str, Any]:
         return result
 
 
+class MiniMaxDuplexTTSOptions(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    key: str = Field(..., min_length=1, repr=False)
+    model: str = Field(..., min_length=1)
+    voice_id: str = Field(..., min_length=1)
+    url: Optional[str] = None
+    speed: Optional[float] = None
+    vol: Optional[float] = None
+    pitch: Optional[float] = None
+    emotion: Optional[str] = None
+    latex_read: Optional[bool] = None
+    sample_rate: Optional[int] = None
+    audio_format: Optional[str] = None
+    channels: Optional[int] = None
+    pronunciation_dict: Optional[Dict[str, Any]] = None
+    voice_modify: Optional[Dict[str, Any]] = None
+    language_boost: Optional[str] = None
+    text_normalization_mode: Optional[str] = None
+    subtitle_enable: Optional[bool] = None
+    subtitle_type: Optional[str] = None
+    continuous_sound: Optional[bool] = None
+    session_id: Optional[str] = None
+    additional_params: Optional[Dict[str, Any]] = None
+    skip_patterns: Optional[List[int]] = None
+
+
+class MiniMaxDuplexTTS(MiniMaxDuplexTTSOptions, _BaseTTSCompat):
+    """MiniMax duplex streaming TTS, currently available only in mainland China."""
+
+    def to_config(self) -> Dict[str, Any]:
+        params: Dict[str, Any] = dict(self.additional_params or {})
+        voice_setting: Dict[str, Any] = {"voice_id": self.voice_id}
+        for field in ("speed", "vol", "pitch", "emotion", "latex_read"):
+            value = getattr(self, field)
+            if value is not None:
+                voice_setting[field] = value
+        params.update({"key": self.key, "model": self.model, "voice_setting": voice_setting})
+        audio_setting: Dict[str, Any] = {}
+        for field, wire_field in (("sample_rate", "sample_rate"), ("audio_format", "format"), ("channels", "channels")):
+            value = getattr(self, field)
+            if value is not None:
+                audio_setting[wire_field] = value
+        if audio_setting:
+            params["audio_setting"] = audio_setting
+        for field in (
+            "url", "pronunciation_dict", "voice_modify", "language_boost", "text_normalization_mode",
+            "subtitle_enable", "subtitle_type", "continuous_sound", "session_id",
+        ):
+            value = getattr(self, field)
+            if value is not None:
+                params[field] = value
+        config: Dict[str, Any] = {"vendor": "minimax_duplex", "params": params}
+        if self.skip_patterns is not None:
+            config["skip_patterns"] = self.skip_patterns
+        return config
+
+
 class MiniMaxTTSOptions(BaseModel):
     model_config = ConfigDict(extra="forbid")