camb-sdk 1.5.1__tar.gz → 1.5.7__tar.gz

{camb_sdk-1.5.1 → camb_sdk-1.5.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: camb-sdk
-Version: 1.5.1
+Version: 1.5.7
 Summary: The official Python SDK for the Camb.ai API
 Author-email: "Camb.ai" <support@camb.ai>
 Classifier: Programming Language :: Python :: 3
@@ -19,9 +19,16 @@ Dynamic: requires-python
 
 # Camb.ai Python SDK
 
+<div id="top" align="center">
+
+   ![Banner](assets/banner5_720.jpg)
+   <h3>
+   <a href="https://camb.ai/"> Camb AI Website </a></h3>
+
 [![PyPI version](https://img.shields.io/pypi/v/camb-sdk.svg?style=flat-square)](https://pypi.org/project/camb-sdk/)  
 [![License](https://img.shields.io/pypi/l/camb-sdk.svg?style=flat-square)](https://github.com/Camb-ai/cambai-python-sdk/blob/main/LICENSE)  
 [![Build status](https://github.com/Camb-ai/cambai-python-sdk/actions/workflows/python.yml/badge.svg)](https://github.com/Camb-ai/cambai-python-sdk/actions/workflows/python.yml)
+</div>
 
 
 The official Python SDK for interacting with Camb AI's powerful voice and audio generation APIs. Create expressive speech, unique voices, and rich soundscapes with just a few lines of Python.
@@ -62,12 +69,57 @@ client = CambAI(api_key="YOUR_CAMB_API_KEY")
 async_client = AsyncCambAI(api_key="YOUR_CAMB_API_KEY")
 ```
 
+
+### Client with Specific MARS Pro Provider (e.g. Vertex, Baseten) 
+#### Baseten
+To deploy the model go to models from baseten example: https://www.baseten.co/library/mars6/ and deploy then perform setup like below
+```python
+client_baseten = CambAI(
+    tts_provider="baseten",
+    provider_params={
+        "api_key": "YOUR_BASETEN_API_KEY",
+        "mars_url": "YOUR_BASETEN_URL"
+    }
+)
+
+# Call TTS with Baseten
+client_baseten.text_to_speech.tts(
+    text="Hello World and my dear friends",
+    language="en-us",
+    speech_model="mars-flash",
+    request_options={
+        "additional_body_parameters": {
+            "reference_audio": base64.b64encode(open("audio.wav", "rb").read()).decode('utf-8'),  # also support public/signed urls
+            "reference_language": "en-us"  # required
+        },
+        "timeout_in_seconds": 300
+    }
+)
+```
+
+#### Vertex Support (In Progress)
+```python
+client_with_provider = CambAI(
+    tts_provider="vertex",
+    provider_params={"project_id": "my-project", "location": "us-central1"}
+)
+```
+
 ## 🚀 Getting Started: Examples
+NOTE: For more examples and full ready to run files refer to the examples/ directory.
 
 ### 1. Text-to-Speech (TTS)
 
 Convert text into spoken audio using one of Camb AI's high-quality voices.
 
+### Supported Models & Sample Rates
+
+| Model Name | Sample Rate | Description |
+| :--- | :--- | :--- |
+| **mars-pro** | **48kHz** | High-fidelity, professional-grade speech synthesis. Ideal for long-form content and dubbing. |
+| **mars-instruct** | **22.05kHz** | optimized for instruction-following and nuance control. |
+| **mars-flash** | **22.05kHz** | Low-latency model optimized for real-time applications and conversational AI. |
+
 #### a) Get an Audio URL or Save to File
 
 ```python
@@ -81,7 +133,7 @@ response = client.text_to_speech.tts(
     text="Hello from Camb AI! This is a test of our Text-to-Speech API.",
     voice_id=20303,  # Example voice ID, get from client.voice_cloning.list_voices()
     language="en-us",
-    speech_model="mars-8",  # options: mars-8, mars-8-flash, mars-8-instruct, auto
+    speech_model="mars-pro",  # options: mars-pro, mars-flash, mars-instruct, auto
     output_configuration=StreamTtsOutputConfiguration(
         format="mp3"
     )
@@ -106,7 +158,7 @@ async def main():
     response = async_client.text_to_speech.tts(
         text="Hello, this is a test of the text to audio streaming capabilities.",
         language="en-us",
-        speech_model="mars-8",  # options: mars-8, mars-8-flash, mars-8-instruct, auto
+        speech_model="mars-pro",  # options: mars-pro, mars-flash, mars-instruct, auto
         voice_id=147319,
         output_configuration=StreamTtsOutputConfiguration(
             format="mp3"
@@ -118,7 +170,23 @@ async def main():
 asyncio.run(main())
 ```
 
-#### c) List Available Voices
+#### c) Using Mars Flash (Low Latency)
+
+For applications requiring faster responses, switch to `mars-flash` (22.05kHz).
+
+```python
+response = client.text_to_speech.tts(
+    text="Hey! I can respond much faster.",
+    language="en-us",
+    speech_model="mars-flash",
+    voice_id=<id>,
+    output_configuration=StreamTtsOutputConfiguration(
+        format="wav"
+    )
+)
+```
+
+#### d) List Available Voices
 
 You can list available voices to find a voice_id that suits your needs:
 
@@ -126,7 +194,7 @@ You can list available voices to find a voice_id that suits your needs:
 voices = client.voice_cloning.list_voices()
 print(f"Found {len(voices)} voices:")
 for voice in voices[:5]:  # Print first 5 as an example
-    print(f"  - ID: {voice.id}, Name: {voice.voice_name}, Gender: {voice.gender}, Language: {voice.language}")
+    print(f"  - ID: {voice["id"]}, Name: {voice["voice_name"]}, Gender: {voice["gender"]}, Language: {voice["language"]}")
 ```
 
 ### 2. Text-to-Voice (Generative Voice)

{camb_sdk-1.5.1 → camb_sdk-1.5.7}/README.md

@@ -1,8 +1,15 @@
 # Camb.ai Python SDK
 
+<div id="top" align="center">
+
+   ![Banner](assets/banner5_720.jpg)
+   <h3>
+   <a href="https://camb.ai/"> Camb AI Website </a></h3>
+
 [![PyPI version](https://img.shields.io/pypi/v/camb-sdk.svg?style=flat-square)](https://pypi.org/project/camb-sdk/)  
 [![License](https://img.shields.io/pypi/l/camb-sdk.svg?style=flat-square)](https://github.com/Camb-ai/cambai-python-sdk/blob/main/LICENSE)  
 [![Build status](https://github.com/Camb-ai/cambai-python-sdk/actions/workflows/python.yml/badge.svg)](https://github.com/Camb-ai/cambai-python-sdk/actions/workflows/python.yml)
+</div>
 
 
 The official Python SDK for interacting with Camb AI's powerful voice and audio generation APIs. Create expressive speech, unique voices, and rich soundscapes with just a few lines of Python.
@@ -43,12 +50,57 @@ client = CambAI(api_key="YOUR_CAMB_API_KEY")
 async_client = AsyncCambAI(api_key="YOUR_CAMB_API_KEY")
 ```
 
+
+### Client with Specific MARS Pro Provider (e.g. Vertex, Baseten) 
+#### Baseten
+To deploy the model go to models from baseten example: https://www.baseten.co/library/mars6/ and deploy then perform setup like below
+```python
+client_baseten = CambAI(
+    tts_provider="baseten",
+    provider_params={
+        "api_key": "YOUR_BASETEN_API_KEY",
+        "mars_url": "YOUR_BASETEN_URL"
+    }
+)
+
+# Call TTS with Baseten
+client_baseten.text_to_speech.tts(
+    text="Hello World and my dear friends",
+    language="en-us",
+    speech_model="mars-flash",
+    request_options={
+        "additional_body_parameters": {
+            "reference_audio": base64.b64encode(open("audio.wav", "rb").read()).decode('utf-8'),  # also support public/signed urls
+            "reference_language": "en-us"  # required
+        },
+        "timeout_in_seconds": 300
+    }
+)
+```
+
+#### Vertex Support (In Progress)
+```python
+client_with_provider = CambAI(
+    tts_provider="vertex",
+    provider_params={"project_id": "my-project", "location": "us-central1"}
+)
+```
+
 ## 🚀 Getting Started: Examples
+NOTE: For more examples and full ready to run files refer to the examples/ directory.
 
 ### 1. Text-to-Speech (TTS)
 
 Convert text into spoken audio using one of Camb AI's high-quality voices.
 
+### Supported Models & Sample Rates
+
+| Model Name | Sample Rate | Description |
+| :--- | :--- | :--- |
+| **mars-pro** | **48kHz** | High-fidelity, professional-grade speech synthesis. Ideal for long-form content and dubbing. |
+| **mars-instruct** | **22.05kHz** | optimized for instruction-following and nuance control. |
+| **mars-flash** | **22.05kHz** | Low-latency model optimized for real-time applications and conversational AI. |
+
 #### a) Get an Audio URL or Save to File
 
 ```python
@@ -62,7 +114,7 @@ response = client.text_to_speech.tts(
     text="Hello from Camb AI! This is a test of our Text-to-Speech API.",
     voice_id=20303,  # Example voice ID, get from client.voice_cloning.list_voices()
     language="en-us",
-    speech_model="mars-8",  # options: mars-8, mars-8-flash, mars-8-instruct, auto
+    speech_model="mars-pro",  # options: mars-pro, mars-flash, mars-instruct, auto
     output_configuration=StreamTtsOutputConfiguration(
         format="mp3"
     )
@@ -87,7 +139,7 @@ async def main():
     response = async_client.text_to_speech.tts(
         text="Hello, this is a test of the text to audio streaming capabilities.",
         language="en-us",
-        speech_model="mars-8",  # options: mars-8, mars-8-flash, mars-8-instruct, auto
+        speech_model="mars-pro",  # options: mars-pro, mars-flash, mars-instruct, auto
         voice_id=147319,
         output_configuration=StreamTtsOutputConfiguration(
             format="mp3"
@@ -99,7 +151,23 @@ async def main():
 asyncio.run(main())
 ```
 
-#### c) List Available Voices
+#### c) Using Mars Flash (Low Latency)
+
+For applications requiring faster responses, switch to `mars-flash` (22.05kHz).
+
+```python
+response = client.text_to_speech.tts(
+    text="Hey! I can respond much faster.",
+    language="en-us",
+    speech_model="mars-flash",
+    voice_id=<id>,
+    output_configuration=StreamTtsOutputConfiguration(
+        format="wav"
+    )
+)
+```
+
+#### d) List Available Voices
 
 You can list available voices to find a voice_id that suits your needs:
 
@@ -107,7 +175,7 @@ You can list available voices to find a voice_id that suits your needs:
 voices = client.voice_cloning.list_voices()
 print(f"Found {len(voices)} voices:")
 for voice in voices[:5]:  # Print first 5 as an example
-    print(f"  - ID: {voice.id}, Name: {voice.voice_name}, Gender: {voice.gender}, Language: {voice.language}")
+    print(f"  - ID: {voice["id"]}, Name: {voice["voice_name"]}, Gender: {voice["gender"]}, Language: {voice["language"]}")
 ```
 
 ### 2. Text-to-Voice (Generative Voice)

{camb_sdk-1.5.1 → camb_sdk-1.5.7}/camb/client.py

@@ -8,6 +8,7 @@ import httpx
 from .core.client_wrapper import AsyncClientWrapper, SyncClientWrapper
 from .core.request_options import RequestOptions
 from .environment import CambApiEnvironment
+from .types.tts_provider import TtsProvider
 from .raw_client import AsyncRawCambApi, RawCambApi
 
 if typing.TYPE_CHECKING:
@@ -28,7 +29,6 @@ if typing.TYPE_CHECKING:
     from .translated_tts.client import AsyncTranslatedTtsClient, TranslatedTtsClient
     from .translation.client import AsyncTranslationClient, TranslationClient
     from .voice_cloning.client import AsyncVoiceCloningClient, VoiceCloningClient
-    from .listen.client import AsyncListenClient, ListenClient
 
 def save_stream_to_file(stream: typing.Iterable[bytes], filename: str) -> None:
     """Saves a byte stream to a file.
@@ -103,12 +103,17 @@ class CambAI:
         *,
         base_url: typing.Optional[str] = None,
         environment: CambApiEnvironment = CambApiEnvironment.DEFAULT,
-        api_key: str,
+        api_key: typing.Optional[str] = None,
         headers: typing.Optional[typing.Dict[str, str]] = None,
         timeout: typing.Optional[float] = None,
         follow_redirects: typing.Optional[bool] = True,
         httpx_client: typing.Optional[httpx.Client] = None,
+        tts_provider: typing.Optional[TtsProvider] = None,
+        provider_params: typing.Optional[typing.Dict[str, typing.Any]] = None,
     ):
+        if api_key is None and (tts_provider is None or provider_params is None):
+            raise ValueError("Please provide either 'api_key' or both 'tts_provider' and 'provider_params'.")
+            
         _defaulted_timeout = (
             timeout if timeout is not None else 60 if httpx_client is None else httpx_client.timeout.read
         )
@@ -122,6 +127,8 @@ class CambAI:
             if follow_redirects is not None
             else httpx.Client(timeout=_defaulted_timeout),
             timeout=_defaulted_timeout,
+            tts_provider=tts_provider,
+            provider_params=provider_params,
         )
         self._raw_client = RawCambApi(client_wrapper=self._client_wrapper)
         self._audio_separation: typing.Optional[AudioSeparationClient] = None
@@ -141,7 +148,6 @@ class CambAI:
         self._dictionaries: typing.Optional[DictionariesClient] = None
         self._project_setup: typing.Optional[ProjectSetupClient] = None
         self._deprecated_streaming: typing.Optional[DeprecatedStreamingClient] = None
-        self._listen: typing.Optional[ListenClient] = None
 
     @property
     def with_raw_response(self) -> RawCambApi:
@@ -364,14 +370,6 @@ class CambAI:
             self._deprecated_streaming = DeprecatedStreamingClient(client_wrapper=self._client_wrapper)
         return self._deprecated_streaming
 
-    @property
-    def listen(self):
-        if self._listen is None:
-            from .listen.client import ListenClient  # noqa: E402
-
-            self._listen = ListenClient(client_wrapper=self._client_wrapper)
-        return self._listen
-
 
 class AsyncCambAI:
     """
@@ -418,12 +416,17 @@ class AsyncCambAI:
         *,
         base_url: typing.Optional[str] = None,
         environment: CambApiEnvironment = CambApiEnvironment.DEFAULT,
-        api_key: str,
+        api_key: typing.Optional[str] = None,
         headers: typing.Optional[typing.Dict[str, str]] = None,
         timeout: typing.Optional[float] = None,
         follow_redirects: typing.Optional[bool] = True,
         httpx_client: typing.Optional[httpx.AsyncClient] = None,
+        tts_provider: typing.Optional[TtsProvider] = None,
+        provider_params: typing.Optional[typing.Dict[str, typing.Any]] = None,
     ):
+        if api_key is None and (tts_provider is None or provider_params is None):
+            raise ValueError("Please provide either 'api_key' or both 'tts_provider' and 'provider_params'.")
+
         _defaulted_timeout = (
             timeout if timeout is not None else 60 if httpx_client is None else httpx_client.timeout.read
         )
@@ -437,6 +440,8 @@ class AsyncCambAI:
             if follow_redirects is not None
             else httpx.AsyncClient(timeout=_defaulted_timeout),
             timeout=_defaulted_timeout,
+            tts_provider=tts_provider,
+            provider_params=provider_params,
         )
         self._raw_client = AsyncRawCambApi(client_wrapper=self._client_wrapper)
         self._audio_separation: typing.Optional[AsyncAudioSeparationClient] = None
@@ -456,7 +461,6 @@ class AsyncCambAI:
         self._dictionaries: typing.Optional[AsyncDictionariesClient] = None
         self._project_setup: typing.Optional[AsyncProjectSetupClient] = None
         self._deprecated_streaming: typing.Optional[AsyncDeprecatedStreamingClient] = None
-        self._listen: typing.Optional[AsyncListenClient] = None
 
     @property
     def with_raw_response(self) -> AsyncRawCambApi:
@@ -703,14 +707,6 @@ class AsyncCambAI:
             self._deprecated_streaming = AsyncDeprecatedStreamingClient(client_wrapper=self._client_wrapper)
         return self._deprecated_streaming
 
-    @property
-    def listen(self):
-        if self._listen is None:
-            from .listen.client import AsyncListenClient  # noqa: E402
-
-            self._listen = AsyncListenClient(client_wrapper=self._client_wrapper)
-        return self._listen
-
 
 def _get_base_url(*, base_url: typing.Optional[str] = None, environment: CambApiEnvironment) -> str:
     if base_url is not None:

{camb_sdk-1.5.1 → camb_sdk-1.5.7}/camb/core/client_wrapper.py

@@ -3,6 +3,7 @@
 import typing
 
 import httpx
+from ..types.tts_provider import TtsProvider
 from .http_client import AsyncHttpClient, HttpClient
 
 
@@ -10,21 +11,30 @@ class BaseClientWrapper:
     def __init__(
         self,
         *,
-        api_key: str,
+        api_key: typing.Optional[str] = None,
         headers: typing.Optional[typing.Dict[str, str]] = None,
         base_url: str,
         timeout: typing.Optional[float] = None,
+        tts_provider: typing.Optional[TtsProvider] = None,
+        provider_params: typing.Optional[typing.Dict[str, typing.Any]] = None,
     ):
         self.api_key = api_key
         self._headers = headers
         self._base_url = base_url
         self._timeout = timeout
+        self.tts_provider = tts_provider
+        self.provider_params = provider_params
 
     def get_headers(self) -> typing.Dict[str, str]:
         headers: typing.Dict[str, str] = {
+            "X-Fern-Language": "Python",
             **(self.get_custom_headers() or {}),
         }
-        headers["x-api-key"] = self.api_key
+        if self.api_key is not None:
+            headers["x-api-key"] = self.api_key
+        if self.tts_provider:
+             headers["tts_provider"] = self.tts_provider
+        # provider_params are not automatically added to headers
         return headers
 
     def get_custom_headers(self) -> typing.Optional[typing.Dict[str, str]]:
@@ -41,13 +51,22 @@ class SyncClientWrapper(BaseClientWrapper):
     def __init__(
         self,
         *,
-        api_key: str,
+        api_key: typing.Optional[str] = None,
         headers: typing.Optional[typing.Dict[str, str]] = None,
         base_url: str,
         timeout: typing.Optional[float] = None,
         httpx_client: httpx.Client,
+        tts_provider: typing.Optional[TtsProvider] = None,
+        provider_params: typing.Optional[typing.Dict[str, typing.Any]] = None,
     ):
-        super().__init__(api_key=api_key, headers=headers, base_url=base_url, timeout=timeout)
+        super().__init__(
+            api_key=api_key, 
+            headers=headers, 
+            base_url=base_url, 
+            timeout=timeout, 
+            tts_provider=tts_provider, 
+            provider_params=provider_params
+        )
         self.httpx_client = HttpClient(
             httpx_client=httpx_client,
             base_headers=self.get_headers,
@@ -60,14 +79,23 @@ class AsyncClientWrapper(BaseClientWrapper):
     def __init__(
         self,
         *,
-        api_key: str,
+        api_key: typing.Optional[str] = None,
         headers: typing.Optional[typing.Dict[str, str]] = None,
         base_url: str,
         timeout: typing.Optional[float] = None,
         async_token: typing.Optional[typing.Callable[[], typing.Awaitable[str]]] = None,
         httpx_client: httpx.AsyncClient,
+        tts_provider: typing.Optional[TtsProvider] = None,
+        provider_params: typing.Optional[typing.Dict[str, typing.Any]] = None,
     ):
-        super().__init__(api_key=api_key, headers=headers, base_url=base_url, timeout=timeout)
+        super().__init__(
+            api_key=api_key, 
+            headers=headers, 
+            base_url=base_url, 
+            timeout=timeout, 
+            tts_provider=tts_provider, 
+            provider_params=provider_params
+        )
         self._async_token = async_token
         self.httpx_client = AsyncHttpClient(
             httpx_client=httpx_client,

{camb_sdk-1.5.1 → camb_sdk-1.5.7}/camb/environment.py

@@ -4,5 +4,4 @@ import enum
 
 
 class CambApiEnvironment(enum.Enum):
-    # DEFAULT = "https://knative-dev-client.camb.ai/apis"
     DEFAULT = "https://client.camb.ai/apis"

camb_sdk-1.5.7/camb/text_to_speech/baseten.py

@@ -0,0 +1,214 @@
+import contextlib
+import typing
+
+import httpx
+from ..core.client_wrapper import AsyncClientWrapper, SyncClientWrapper
+from ..core.http_response import AsyncHttpResponse, HttpResponse
+from ..core.request_options import RequestOptions
+from ..core.serialization import convert_and_respect_annotation_metadata
+from ..types.stream_tts_inference_options import StreamTtsInferenceOptions
+from ..types.stream_tts_output_configuration import StreamTtsOutputConfiguration
+from ..types.stream_tts_voice_settings import StreamTtsVoiceSettings
+from .types.create_stream_tts_request_payload_language import CreateStreamTtsRequestPayloadLanguage
+from .types.create_stream_tts_request_payload_speech_model import CreateStreamTtsRequestPayloadSpeechModel
+
+OMIT = typing.cast(typing.Any, ...)
+
+
+@contextlib.contextmanager
+def baseten_tts(
+    client_wrapper: SyncClientWrapper,
+    *,
+    text: str,
+    language: CreateStreamTtsRequestPayloadLanguage,
+    voice_id: typing.Optional[int] = OMIT,
+    speech_model: typing.Optional[CreateStreamTtsRequestPayloadSpeechModel] = OMIT,
+    user_instructions: typing.Optional[str] = OMIT,
+    enhance_named_entities_pronunciation: typing.Optional[bool] = OMIT,
+    output_configuration: typing.Optional[StreamTtsOutputConfiguration] = OMIT,
+    voice_settings: typing.Optional[StreamTtsVoiceSettings] = OMIT,
+    inference_options: typing.Optional[StreamTtsInferenceOptions] = OMIT,
+    request_options: typing.Optional[RequestOptions] = None,
+) -> typing.Iterator[httpx.Response]:
+    # Retrieve API key from provider_params
+    provider_params = client_wrapper.provider_params or {}
+    api_key = provider_params.get("api_key", "")
+    mars_pro_url = provider_params.get("mars_pro_url") or provider_params.get("mars_url")
+    if not mars_pro_url:
+        raise ValueError("mars_url is required for using Baseten as provider")
+
+    headers = {
+        "Authorization": f"Api-Key {api_key}" if api_key else "",
+        "Content-Type": "application/json",
+    }
+    # Construct Payload
+    # 1. Basic Fields
+    payload = {
+        "text": text,
+        "language": str(language).lower().replace("_", "-"),
+        "output_format": "wav",
+        "stream": True,
+        "apply_ner_nlp": False,
+    }
+
+    # 2. Output Configuration
+    if output_configuration and output_configuration is not OMIT:
+        if output_configuration.format:
+            payload["output_format"] = str(output_configuration.format)
+
+    # 3. Voice Settings
+    if voice_settings and voice_settings is not OMIT:
+        if voice_settings.enhance_reference_audio_quality is not None:
+             payload["apply_ref_mpsenet"] = voice_settings.enhance_reference_audio_quality
+        if voice_settings.maintain_source_accent:
+             payload["accent_nudge"] = 0.8
+
+    # 4. Inference Options
+    if inference_options and inference_options is not OMIT:
+        if inference_options.temperature is not None:
+             payload["temperature"] = inference_options.temperature
+        if inference_options.inference_steps is not None:
+             payload["inference_steps"] = inference_options.inference_steps
+            
+        if inference_options.speaker_similarity is not None:
+             # Formula from user snippet:
+             s = max(0.0, min(0.7, inference_options.speaker_similarity))
+             payload["campp_speaker_nudge"] = 1.5 * (1 - s / 0.7)
+
+    # 5. Extract additional params (reference_audio, reference_language) from request_options if present
+    #    This allows passing 'reference_audio' without breaking the explicit signature for now.
+    extra_body = {}
+    if request_options and request_options.get("additional_body_parameters"):
+        extra_body = request_options.get("additional_body_parameters")
+    
+    if "reference_audio" not in extra_body:
+        raise ValueError("reference_audio is required in additional_body_parameters for Baseten provider")
+    if "reference_language" not in extra_body:
+        raise ValueError("reference_language is required in additional_body_parameters for Baseten provider")
+
+    payload["reference_language"] = extra_body["reference_language"]
+    payload["audio_ref"] = extra_body["reference_audio"]
+    payload["reference_audio"] = extra_body["reference_audio"]
+
+    timeout = None
+    if request_options and request_options.get("timeout_in_seconds") is not None:
+        timeout = request_options.get("timeout_in_seconds")
+
+    # Use the raw httpx client to avoid SDK wrapper injecting unwanted headers/params
+    # that might interfere with Baseten's strict endpoint.
+    with client_wrapper.httpx_client.httpx_client.stream(
+        "POST",
+        mars_pro_url,
+        json=payload,
+        headers=headers,
+        timeout=timeout
+    ) as _response:
+        # Check status manually since we bypassed the wrapper's check
+        if not (200 <= _response.status_code < 300):
+            # Try to read error body
+            _response.read()
+            raise Exception(f"Baseten API Error: {_response.status_code} - {_response.text}")
+
+        yield HttpResponse(
+            response=_response,
+            data=(_chunk for _chunk in _response.iter_bytes(chunk_size=None)),
+        )
+
+
+@contextlib.asynccontextmanager
+async def async_baseten_tts(
+    client_wrapper: AsyncClientWrapper,
+    *,
+    text: str,
+    language: CreateStreamTtsRequestPayloadLanguage,
+    voice_id: typing.Optional[int] = OMIT,
+    speech_model: typing.Optional[CreateStreamTtsRequestPayloadSpeechModel] = OMIT,
+    user_instructions: typing.Optional[str] = OMIT,
+    enhance_named_entities_pronunciation: typing.Optional[bool] = OMIT,
+    output_configuration: typing.Optional[StreamTtsOutputConfiguration] = OMIT,
+    voice_settings: typing.Optional[StreamTtsVoiceSettings] = OMIT,
+    inference_options: typing.Optional[StreamTtsInferenceOptions] = OMIT,
+    request_options: typing.Optional[RequestOptions] = None,
+) -> typing.AsyncIterator[AsyncHttpResponse[typing.AsyncIterator[bytes]]]:
+    # Retrieve API key from provider_params
+    provider_params = client_wrapper.provider_params or {}
+    api_key = provider_params.get("api_key", "")
+    mars_pro_url = provider_params.get("mars_pro_url") or provider_params.get("mars_url")
+    if not mars_pro_url:
+        raise ValueError("mars_url is required for using Baseten as provider")
+    api_key_header_val = f"Api-Key {api_key}"
+
+    # Construct Payload
+    # 1. Basic Fields
+    payload = {
+        "text": text,
+        "language": str(language).lower().replace("_", "-"),
+        "stream": True,
+        "output_format": "wav",  # Default
+        "apply_ner_nlp": False, # Default based on doc
+    }
+
+    # 2. Output Configuration
+    if output_configuration and output_configuration is not OMIT:
+        if output_configuration.format:
+            payload["output_format"] = str(output_configuration.format)
+
+    # 3. Voice Settings
+    if voice_settings and voice_settings is not OMIT:
+        if voice_settings.enhance_reference_audio_quality is not None:
+             payload["apply_ref_mpsenet"] = voice_settings.enhance_reference_audio_quality
+        if voice_settings.maintain_source_accent:
+             payload["accent_nudge"] = 0.8
+
+    # 4. Inference Options
+    if inference_options and inference_options is not OMIT:
+        if inference_options.temperature is not None:
+             payload["temperature"] = inference_options.temperature
+        if inference_options.inference_steps is not None:
+             payload["inference_steps"] = inference_options.inference_steps
+            
+        if inference_options.speaker_similarity is not None:
+             # Formula from user snippet:
+             s = max(0.0, min(0.7, inference_options.speaker_similarity))
+             payload["campp_speaker_nudge"] = 1.5 * (1 - s / 0.7)
+
+    # 5. Extract additional params (reference_audio, reference_language) from request_options
+    extra_body = {}
+    if request_options and request_options.get("additional_body_parameters"):
+        extra_body = request_options.get("additional_body_parameters")
+    
+    if "reference_audio" not in extra_body:
+        raise ValueError("reference_audio is required in additional_body_parameters for Baseten provider")
+    if "reference_language" not in extra_body:
+        raise ValueError("reference_language is required in additional_body_parameters for Baseten provider")
+
+    payload["reference_language"] = extra_body["reference_language"]
+    payload["audio_ref"] = extra_body["reference_audio"]
+    payload["reference_audio"] = extra_body["reference_audio"]
+
+    timeout = None
+    if request_options and request_options.get("timeout_in_seconds") is not None:
+        timeout = request_options.get("timeout_in_seconds")
+    
+    # Use the raw httpx client to avoid SDK wrapper injecting unwanted headers/params
+    # that might interfere with Baseten's strict endpoint.
+    async with client_wrapper.httpx_client.httpx_client.stream(
+        "POST",
+        mars_pro_url,
+        json=payload,
+        headers={
+            "Authorization": api_key_header_val,
+            "content-type": "application/json",
+        },
+        timeout=timeout
+    ) as _response:
+        # Check status manually since we bypassed the wrapper's check
+        if not (200 <= _response.status_code < 300):
+             # Try to read error body
+             await _response.aread()
+             raise Exception(f"Baseten API Error: {_response.status_code} - {_response.text}")
+
+        yield AsyncHttpResponse(
+            response=_response,
+            data=(_chunk async for _chunk in _response.aiter_bytes(chunk_size=None)),
+        )