cartesia 0.0.3__py2.py3-none-any.whl → 0.0.5__py2.py3-none-any.whl

cartesia/__init__.py

@@ -1,3 +1,3 @@
-from cartesia.tts import CartesiaTTS
+from cartesia.tts import AsyncCartesiaTTS, CartesiaTTS
 
-__all__ = ["CartesiaTTS"]
+__all__ = ["CartesiaTTS", "AsyncCartesiaTTS"]

cartesia/tts.py

@@ -1,16 +1,29 @@
+import asyncio
 import base64
 import json
 import os
 import uuid
-from typing import Any, Dict, Generator, List, Optional, TypedDict, Union
+from types import TracebackType
+from typing import Any, AsyncGenerator, Dict, Generator, List, Optional, Tuple, TypedDict, Union
 
+import aiohttp
+import httpx
+import logging
 import requests
 from websockets.sync.client import connect
 
+from cartesia.utils import retry_on_connection_error, retry_on_connection_error_async
+
 DEFAULT_MODEL_ID = "genial-planet-1346"
 DEFAULT_BASE_URL = "api.cartesia.ai"
 DEFAULT_API_VERSION = "v0"
+DEFAULT_TIMEOUT = 30  # seconds
+DEFAULT_NUM_CONNECTIONS = 10  # connections per client
+
+BACKOFF_FACTOR = 1
+MAX_RETRIES = 3
 
+logger = logging.getLogger(__name__)
 
 class AudioOutput(TypedDict):
     audio: bytes
@@ -27,14 +40,48 @@ class VoiceMetadata(TypedDict):
     embedding: Optional[Embedding]
 
 
+def update_buffer(buffer: str, chunk_bytes: bytes) -> Tuple[str, List[Dict[str, Any]]]:
+    buffer += chunk_bytes.decode("utf-8")
+    outputs = []
+    while "{" in buffer and "}" in buffer:
+        start_index = buffer.find("{")
+        end_index = buffer.find("}", start_index)
+        if start_index != -1 and end_index != -1:
+            try:
+                chunk_json = json.loads(buffer[start_index : end_index + 1])
+                audio = base64.b64decode(chunk_json["data"])
+                outputs.append({"audio": audio, "sampling_rate": chunk_json["sampling_rate"]})
+                buffer = buffer[end_index + 1 :]
+            except json.JSONDecodeError:
+                break
+    return buffer, outputs
+
+
+def convert_response(response: Dict[str, any], include_context_id: bool) -> Dict[str, Any]:
+    audio = base64.b64decode(response["data"])
+
+    optional_kwargs = {}
+    if include_context_id:
+        optional_kwargs["context_id"] = response["context_id"]
+
+    return {
+        "audio": audio,
+        "sampling_rate": response["sampling_rate"],
+        **optional_kwargs,
+    }
+
+
 class CartesiaTTS:
     """The client for Cartesia's text-to-speech library.
 
     This client contains methods to interact with the Cartesia text-to-speech API.
-    The API offers
+    The client can be used to retrieve available voices, compute new voice embeddings,
+    and generate speech from text.
 
-    Examples:
+    The client also supports generating audio using a websocket for lower latency.
+    To enable interrupt handling along the websocket, set `experimental_ws_handle_interrupts=True`.
 
+    Examples:
         >>> client = CartesiaTTS()
 
         # Load available voices and their metadata (excluding the embeddings).
@@ -55,19 +102,21 @@ class CartesiaTTS:
         ...     audio, sr = audio_chunk["audio"], audio_chunk["sampling_rate"]
     """
 
-    def __init__(self, *, api_key: str = None):
-        """
-        Args:
-            api_key: The API key to use for authorization.
-                If not specified, the API key will be read from the environment variable
-                `CARTESIA_API_KEY`.
+    def __init__(self, *, api_key: str = None, experimental_ws_handle_interrupts: bool = False):
+        """Args:
+        api_key: The API key to use for authorization.
+            If not specified, the API key will be read from the environment variable
+            `CARTESIA_API_KEY`.
+        experimental_ws_handle_interrupts: Whether to handle interrupts when generating
+            audio using the websocket. This is an experimental feature and may have bugs
+            or be deprecated in the future.
         """
         self.base_url = os.environ.get("CARTESIA_BASE_URL", DEFAULT_BASE_URL)
         self.api_key = api_key or os.environ.get("CARTESIA_API_KEY")
         self.api_version = os.environ.get("CARTESIA_API_VERSION", DEFAULT_API_VERSION)
         self.headers = {"X-API-Key": self.api_key, "Content-Type": "application/json"}
         self.websocket = None
-        self.refresh_websocket()
+        self.experimental_ws_handle_interrupts = experimental_ws_handle_interrupts
 
     def get_voices(self, skip_embeddings: bool = True) -> Dict[str, VoiceMetadata]:
         """Returns a mapping from voice name -> voice metadata.
@@ -100,9 +149,14 @@ class CartesiaTTS:
             >>> audio = client.generate(transcript="Hello world!", voice=embedding)
         """
         params = {"select": "id, name, description"} if skip_embeddings else None
-        response = requests.get(f"{self._http_url()}/voices", headers=self.headers, params=params)
+        response = httpx.get(
+            f"{self._http_url()}/voices",
+            headers=self.headers,
+            params=params,
+            timeout=DEFAULT_TIMEOUT,
+        )
 
-        if response.status_code != 200:
+        if not response.is_success:
             raise ValueError(f"Failed to get voices. Error: {response.text}")
 
         voices = response.json()
@@ -112,6 +166,7 @@ class CartesiaTTS:
                 voice["embedding"] = json.loads(voice["embedding"])
         return {voice["name"]: voice for voice in voices}
 
+    @retry_on_connection_error(max_retries=MAX_RETRIES, backoff_factor=BACKOFF_FACTOR, logger=logger)
     def get_voice_embedding(
         self, *, voice_id: str = None, filepath: str = None, link: str = None
     ) -> Embedding:
@@ -134,20 +189,20 @@ class CartesiaTTS:
 
         if voice_id:
             url = f"{self._http_url()}/voices/embedding/{voice_id}"
-            response = requests.get(url, headers=self.headers)
+            response = httpx.get(url, headers=self.headers, timeout=DEFAULT_TIMEOUT)
         elif filepath:
             url = f"{self._http_url()}/voices/clone/clip"
             files = {"clip": open(filepath, "rb")}
             headers = self.headers.copy()
             # The default content type of JSON is incorrect for file uploads
             headers.pop("Content-Type")
-            response = requests.post(url, headers=headers, files=files)
+            response = httpx.post(url, headers=headers, files=files, timeout=DEFAULT_TIMEOUT)
         elif link:
             url = f"{self._http_url()}/voices/clone/url"
             params = {"link": link}
-            response = requests.post(url, headers=self.headers, params=params)
+            response = httpx.post(url, headers=self.headers, params=params, timeout=DEFAULT_TIMEOUT)
 
-        if response.status_code != 200:
+        if not response.is_success:
             raise ValueError(
                 f"Failed to clone voice. Status Code: {response.status_code}\n"
                 f"Error: {response.text}"
@@ -165,12 +220,11 @@ class CartesiaTTS:
         Note:
             The connection is synchronous.
         """
-        if self.websocket and not self._is_websocket_closed():
-            self.websocket.close()
-        self.websocket = connect(
-            f"{self._ws_url()}/audio/websocket?api_key={self.api_key}",
-            close_timeout=None,
-        )
+        if self.websocket is None or self._is_websocket_closed():
+            route = "audio/websocket"
+            if self.experimental_ws_handle_interrupts:
+                route = f"experimental/{route}"
+            self.websocket = connect(f"{self._ws_url()}/{route}?api_key={self.api_key}")
 
     def _is_websocket_closed(self):
         return self.websocket.socket.fileno() == -1
@@ -189,29 +243,53 @@ class CartesiaTTS:
         if transcript.strip() == "":
             raise ValueError("`transcript` must be non empty")
 
+    def _generate_request_body(
+        self,
+        *,
+        transcript: str,
+        voice: Embedding,
+        model_id: str,
+        duration: int = None,
+        chunk_time: float = None,
+    ) -> Dict[str, Any]:
+        """Create the request body for a stream request.
+
+        Note that anything that's not provided will use a default if available or be
+        filtered out otherwise.
+        """
+        body = dict(transcript=transcript, model_id=model_id, voice=voice)
+
+        optional_body = dict(
+            duration=duration,
+            chunk_time=chunk_time,
+        )
+        body.update({k: v for k, v in optional_body.items() if v is not None})
+
+        return body
+
     def generate(
         self,
         *,
         transcript: str,
+        voice: Embedding,
+        model_id: str = DEFAULT_MODEL_ID,
         duration: int = None,
         chunk_time: float = None,
-        voice: Embedding = None,
         stream: bool = False,
         websocket: bool = True,
     ) -> Union[AudioOutput, Generator[AudioOutput, None, None]]:
         """Generate audio from a transcript.
 
         Args:
-            transcript: The text to generate audio for.
-            duration: The maximum duration of the audio in seconds.
-            chunk_time: How long each audio segment should be in seconds.
+            transcript (str): The text to generate audio for.
+            voice (Embedding (List[float])): The voice to use for generating audio.
+            duration (int, optional): The maximum duration of the audio in seconds.
+            chunk_time (float, optional): How long each audio segment should be in seconds.
                 This should not need to be adjusted.
-            voice: The voice to use for generating audio.
-                This can either be a voice id (string) or an embedding vector (List[float]).
-            stream: Whether to stream the audio or not.
-                If ``True`` this function returns a generator.
-            websocket: Whether to use a websocket for streaming audio.
-                Using the websocket reduces latency by pre-poning the handshake.
+            stream (bool, optional): Whether to stream the audio or not.
+                If True this function returns a generator. False by default.
+            websocket (bool, optional): Whether to use a websocket for streaming audio.
+                Using the websocket reduces latency by pre-poning the handshake. True by default.
 
         Returns:
             A generator if `stream` is True, otherwise a dictionary.
@@ -221,19 +299,18 @@ class CartesiaTTS:
         """
         self._check_inputs(transcript, duration, chunk_time)
 
-        body = dict(transcript=transcript, model_id=DEFAULT_MODEL_ID)
-
-        optional_body = dict(
-            duration=duration,
-            chunk_time=chunk_time,
-            voice=voice,
+        body = self._generate_request_body(
+            transcript=transcript, 
+            voice=voice, 
+            model_id=model_id,
+            duration=duration, 
+            chunk_time=chunk_time
         )
-        body.update({k: v for k, v in optional_body.items() if v is not None})
 
         if websocket:
             generator = self._generate_ws(body)
         else:
-            generator = self._generate_http(body)
+            generator = self._generate_http_wrapper(body)
 
         if stream:
             return generator
@@ -247,30 +324,32 @@ class CartesiaTTS:
 
         return {"audio": b"".join(chunks), "sampling_rate": sampling_rate}
 
+    @retry_on_connection_error(max_retries=MAX_RETRIES, backoff_factor=BACKOFF_FACTOR, logger=logger)
+    def _generate_http_wrapper(self, body: Dict[str, Any]):
+        """Need to wrap the http generator in a function for the retry decorator to work."""
+        try:
+            for chunk in self._generate_http(body):
+                yield chunk
+        except Exception as e:
+            logger.error(f"Failed to generate audio. {e}")
+            raise e
+
     def _generate_http(self, body: Dict[str, Any]):
         response = requests.post(
             f"{self._http_url()}/audio/stream",
             stream=True,
             data=json.dumps(body),
             headers=self.headers,
+            timeout=(DEFAULT_TIMEOUT, DEFAULT_TIMEOUT),
         )
-        if response.status_code != 200:
+        if not response.ok:
             raise ValueError(f"Failed to generate audio. {response.text}")
 
         buffer = ""
         for chunk_bytes in response.iter_content(chunk_size=None):
-            buffer += chunk_bytes.decode("utf-8")
-            while "{" in buffer and "}" in buffer:
-                start_index = buffer.find("{")
-                end_index = buffer.find("}", start_index)
-                if start_index != -1 and end_index != -1:
-                    try:
-                        chunk_json = json.loads(buffer[start_index : end_index + 1])
-                        audio = base64.b64decode(chunk_json["data"])
-                        yield {"audio": audio, "sampling_rate": chunk_json["sampling_rate"]}
-                        buffer = buffer[end_index + 1 :]
-                    except json.JSONDecodeError:
-                        break
+            buffer, outputs = update_buffer(buffer, chunk_bytes)
+            for output in outputs:
+                yield output
 
         if buffer:
             try:
@@ -280,21 +359,77 @@ class CartesiaTTS:
             except json.JSONDecodeError:
                 pass
 
-    def _generate_ws(self, body: Dict[str, Any]):
+    def _generate_ws(self, body: Dict[str, Any], *, context_id: str = None):
+        """Generate audio using the websocket connection.
+
+        Args:
+            body: The request body.
+            context_id: The context id for the request.
+                The context id must be globally unique for the duration this client exists.
+                If this is provided, the context id that is in the response will
+                also be returned as part of the dict. This is helpful for testing.
+        """
         if not self.websocket or self._is_websocket_closed():
             self.refresh_websocket()
 
-        self.websocket.send(json.dumps({"data": body, "context_id": uuid.uuid4().hex}))
+        include_context_id = bool(context_id)
+        if context_id is None:
+            context_id = uuid.uuid4().hex
+        self.websocket.send(json.dumps({"data": body, "context_id": context_id}))
         try:
-            response = json.loads(self.websocket.recv())
-            while not response["done"]:
-                audio = base64.b64decode(response["data"])
-                # print("timing", time.perf_counter() - start)
-                yield {"audio": audio, "sampling_rate": response["sampling_rate"]}
-
+            while True:
                 response = json.loads(self.websocket.recv())
-        except Exception:
-            raise RuntimeError(f"Failed to generate audio. {response}")
+                if response["done"]:
+                    break
+
+                yield convert_response(response, include_context_id)
+
+                if self.experimental_ws_handle_interrupts:
+                    self.websocket.send(json.dumps({"context_id": context_id}))
+        except GeneratorExit:
+            # The exit is only called when the generator is garbage collected.
+            # It may not be called directly after a break statement.
+            # However, the generator will be automatically cancelled on the next request.
+            if self.experimental_ws_handle_interrupts:
+                self.websocket.send(json.dumps({"context_id": context_id, "action": "cancel"}))
+        except Exception as e:
+            # Close the websocket connection if an error occurs.
+            if self.websocket and not self._is_websocket_closed():
+                self.websocket.close()
+            raise RuntimeError(f"Failed to generate audio. {response}") from e
+        finally:
+            # Ensure the websocket is ultimately closed.
+            if self.websocket and not self._is_websocket_closed():
+                self.websocket.close()
+
+    @retry_on_connection_error(max_retries=MAX_RETRIES, backoff_factor=BACKOFF_FACTOR, logger=logger)
+    def transcribe(self, raw_audio: Union[bytes, str]) -> str:
+        raw_audio_bytes, headers = self.prepare_audio_and_headers(raw_audio)
+        response = httpx.post(
+            f"{self._http_url()}/audio/transcriptions",
+            headers=headers,
+            files={"clip": ("input.wav", raw_audio_bytes)},
+            timeout=DEFAULT_TIMEOUT,
+        )
+
+        if not response.is_success:
+            raise ValueError(f"Failed to transcribe audio. Error: {response.text()}")
+
+        transcript = response.json()
+        return transcript["text"]
+
+
+    def prepare_audio_and_headers(
+        self, raw_audio: Union[bytes, str]
+    ) -> Tuple[bytes, Dict[str, Any]]:
+        if isinstance(raw_audio, str):
+            with open(raw_audio, "rb") as f:
+                raw_audio_bytes = f.read()
+        else:
+            raw_audio_bytes = raw_audio
+        # application/json is not the right content type for this request
+        headers = {k: v for k, v in self.headers.items() if k != "Content-Type"}
+        return raw_audio_bytes, headers
 
     def _http_url(self):
         prefix = "http" if "localhost" in self.base_url else "https"
@@ -304,6 +439,235 @@ class CartesiaTTS:
         prefix = "ws" if "localhost" in self.base_url else "wss"
         return f"{prefix}://{self.base_url}/{self.api_version}"
 
-    def __del__(self):
-        if self.websocket.socket.fileno() > -1:
+    def close(self):
+        if self.websocket and not self._is_websocket_closed():
             self.websocket.close()
+
+    def __del__(self):
+        self.close()
+
+    def __enter__(self):
+        self.refresh_websocket()
+        return self
+
+    def __exit__(
+        self,
+        exc_type: Union[type, None],
+        exc: Union[BaseException, None],
+        exc_tb: Union[TracebackType, None],
+    ):
+        self.close()
+
+
+class AsyncCartesiaTTS(CartesiaTTS):
+    def __init__(self, *, api_key: str = None, experimental_ws_handle_interrupts: bool = False):
+        self._session = None
+        self._loop = None
+        super().__init__(
+            api_key=api_key, experimental_ws_handle_interrupts=experimental_ws_handle_interrupts
+        )
+    
+    async def _get_session(self):
+        current_loop = asyncio.get_event_loop()
+        if self._loop is not current_loop:
+            # If the loop has changed, close the session and create a new one.
+            await self.close()
+        if self._session is None or self._session.closed:
+            timeout = aiohttp.ClientTimeout(total=DEFAULT_TIMEOUT)
+            connector = aiohttp.TCPConnector(limit=DEFAULT_NUM_CONNECTIONS)
+            self._session = aiohttp.ClientSession(
+                timeout=timeout, connector=connector
+            )
+            self._loop = current_loop
+        return self._session
+    
+    async def refresh_websocket(self):
+        """Refresh the websocket connection."""
+        if self.websocket is None or self._is_websocket_closed():
+            route = "audio/websocket"
+            if self.experimental_ws_handle_interrupts:
+                route = f"experimental/{route}"
+            session = await self._get_session()
+            self.websocket = await session.ws_connect(
+                f"{self._ws_url()}/{route}?api_key={self.api_key}"
+            )
+    
+    def _is_websocket_closed(self):
+        return self.websocket.closed
+
+    async def close(self):
+        """This method closes the websocket and the session.
+        
+        It is *strongly* recommended to call this method when you are done using the client.
+        """
+        if self.websocket is not None and not self._is_websocket_closed():
+            await self.websocket.close()
+        if self._session is not None and not self._session.closed:
+            await self._session.close()
+
+    async def generate(
+        self,
+        *,
+        transcript: str,
+        voice: Embedding,
+        model_id: str = DEFAULT_MODEL_ID,
+        duration: int = None,
+        chunk_time: float = None,
+        stream: bool = False,
+        websocket: bool = True,
+    ) -> Union[AudioOutput, AsyncGenerator[AudioOutput, None]]:
+        """Asynchronously generate audio from a transcript.
+        NOTE: This overrides the non-asynchronous generate method from the base class.
+
+        Args:
+            transcript (str): The text to generate audio for.
+            voice (Embedding (List[float])): The voice to use for generating audio.
+            duration (int, optional): The maximum duration of the audio in seconds.
+            chunk_time (float, optional): How long each audio segment should be in seconds.
+                This should not need to be adjusted.
+            stream (bool, optional): Whether to stream the audio or not.
+                If True this function returns a generator. False by default.
+            websocket (bool, optional): Whether to use a websocket for streaming audio.
+                Using the websocket reduces latency by pre-poning the handshake. True by default.
+
+        Returns:
+            A generator if `stream` is True, otherwise a dictionary.
+            Dictionary from both generator and non-generator return types have the following keys:
+                * "audio": The audio as a bytes buffer.
+                * "sampling_rate": The sampling rate of the audio.
+        """
+        self._check_inputs(transcript, duration, chunk_time)
+
+        body = self._generate_request_body(
+            transcript=transcript, 
+            voice=voice,
+            model_id=model_id,
+            duration=duration, 
+            chunk_time=chunk_time
+        )
+
+        if websocket:
+            generator = self._generate_ws(body)
+        else:
+            generator = self._generate_http_wrapper(body)
+
+        if stream:
+            return generator
+
+        chunks = []
+        sampling_rate = None
+        async for chunk in generator:
+            if sampling_rate is None:
+                sampling_rate = chunk["sampling_rate"]
+            chunks.append(chunk["audio"])
+
+        return {"audio": b"".join(chunks), "sampling_rate": sampling_rate}
+
+    @retry_on_connection_error_async(max_retries=MAX_RETRIES, backoff_factor=BACKOFF_FACTOR, logger=logger)
+    async def _generate_http_wrapper(self, body: Dict[str, Any]):
+        """Need to wrap the http generator in a function for the retry decorator to work."""
+        try:
+          async for chunk in self._generate_http(body):
+              yield chunk
+        except Exception as e:
+            logger.error(f"Failed to generate audio. {e}")
+            raise e
+
+    async def _generate_http(self, body: Dict[str, Any]):
+        session = await self._get_session()
+        async with session.post(
+            f"{self._http_url()}/audio/stream", data=json.dumps(body), headers=self.headers
+        ) as response:
+            if not response.ok:
+                raise ValueError(f"Failed to generate audio. {await response.text()}")
+
+            buffer = ""
+            async for chunk_bytes in response.content.iter_any():
+                buffer, outputs = update_buffer(buffer, chunk_bytes)
+                for output in outputs:
+                    yield output
+
+            if buffer:
+                try:
+                    chunk_json = json.loads(buffer)
+                    audio = base64.b64decode(chunk_json["data"])
+                    yield {"audio": audio, "sampling_rate": chunk_json["sampling_rate"]}
+                except json.JSONDecodeError:
+                    pass
+
+    async def _generate_ws(self, body: Dict[str, Any], *, context_id: str = None):
+        include_context_id = bool(context_id)
+        route = "audio/websocket"
+        if self.experimental_ws_handle_interrupts:
+            route = f"experimental/{route}"
+
+        if not self.websocket or self._is_websocket_closed():
+            await self.refresh_websocket()
+
+        ws = self.websocket
+        if context_id is None:
+            context_id = uuid.uuid4().hex
+        await ws.send_json({"data": body, "context_id": context_id})
+        try:
+            response = None
+            while True:
+                response = await ws.receive_json()
+                if response["done"]:
+                    break
+
+                yield convert_response(response, include_context_id)
+
+                if self.experimental_ws_handle_interrupts:
+                    await ws.send_json({"context_id": context_id})
+        except GeneratorExit:
+            # The exit is only called when the generator is garbage collected.
+            # It may not be called directly after a break statement.
+            # However, the generator will be automatically cancelled on the next request.
+            if self.experimental_ws_handle_interrupts:
+                await ws.send_json({"context_id": context_id, "action": "cancel"})
+        except Exception as e:
+            if self.websocket and not self._is_websocket_closed():
+                await self.websocket.close()
+            raise RuntimeError(f"Failed to generate audio. {await response.text()}") from e
+        finally:
+            # Ensure the websocket is ultimately closed.
+            if self.websocket and not self._is_websocket_closed():
+                await self.websocket.close()
+
+    async def transcribe(self, raw_audio: Union[bytes, str]) -> str:
+        raw_audio_bytes, headers = self.prepare_audio_and_headers(raw_audio)
+        data = aiohttp.FormData()
+        data.add_field("clip", raw_audio_bytes, filename="input.wav", content_type="audio/wav")
+        session = await self._get_session()
+
+        async with session.post(
+            f"{self._http_url()}/audio/transcriptions", headers=headers, data=data
+        ) as response:
+            if not response.ok:
+                raise ValueError(f"Failed to transcribe audio. Error: {await response.text()}")
+
+            transcript = await response.json()
+            return transcript["text"]
+        
+    def __del__(self):
+        try:
+            loop = asyncio.get_running_loop()
+        except RuntimeError:
+            loop = None
+
+        if loop is None:
+            asyncio.run(self.close())
+        else:
+            loop.create_task(self.close())
+
+    async def __aenter__(self):
+        await self.refresh_websocket()
+        return self
+
+    async def __aexit__(
+        self,
+        exc_type: Union[type, None],
+        exc: Union[BaseException, None],
+        exc_tb: Union[TracebackType, None],
+    ):
+        await self.close()

cartesia/utils.py

@@ -0,0 +1,65 @@
+import time
+
+from aiohttp.client_exceptions import ServerDisconnectedError
+import asyncio
+from functools import wraps
+from http.client import RemoteDisconnected
+from httpx import TimeoutException
+from requests.exceptions import ConnectionError
+
+def retry_on_connection_error(max_retries=3, backoff_factor=1, logger=None):
+    """Retry a function if a ConnectionError, RemoteDisconnected, ServerDisconnectedError, or TimeoutException occurs.
+
+    Args:
+        max_retries (int): The maximum number of retries.
+        backoff_factor (int): The factor to increase the delay between retries.
+        logger (logging.Logger): The logger to use for logging.
+    """
+    def decorator(func):
+        @wraps(func)
+        def wrapper(*args, **kwargs):
+            retry_count = 0
+            while retry_count < max_retries:
+                try:
+                    return func(*args, **kwargs)
+                except (ConnectionError, RemoteDisconnected, ServerDisconnectedError, TimeoutException) as e:
+                    logger.info(f"Retrying after exception: {e}")
+                    retry_count += 1
+                    if retry_count < max_retries:
+                        delay = backoff_factor * (2 ** (retry_count - 1))
+                        logger.warn(f"Attempt {retry_count + 1}/{max_retries} in {delay} seconds...")
+                        time.sleep(delay)
+                    else:
+                        raise Exception(f"Exception occurred after {max_retries} tries.") from e
+        return wrapper
+    return decorator
+
+def retry_on_connection_error_async(max_retries=3, backoff_factor=1, logger=None):
+    """Retry an asynchronous function if a ConnectionError, RemoteDisconnected, ServerDisconnectedError, or TimeoutException occurs.
+
+    Args:
+        max_retries (int): The maximum number of retries.
+        backoff_factor (int): The factor to increase the delay between retries.
+        logger (logging.Logger): The logger to use for logging.
+    """
+    def decorator(func):
+        @wraps(func)
+        async def wrapper(*args, **kwargs):
+            retry_count = 0
+            while retry_count < max_retries:
+                try:
+                    async for chunk in func(*args, **kwargs):
+                        yield chunk
+                    # If the function completes without raising an exception return
+                    return
+                except (ConnectionError, RemoteDisconnected, ServerDisconnectedError, TimeoutException) as e:
+                    logger.info(f"Retrying after exception: {e}")
+                    retry_count += 1
+                    if retry_count < max_retries:
+                        delay = backoff_factor * (2 ** (retry_count - 1))
+                        logger.warn(f"Attempt {retry_count + 1}/{max_retries} in {delay} seconds...")
+                        await asyncio.sleep(delay)
+                    else:
+                        raise Exception(f"Exception occurred after {max_retries} tries.") from e
+        return wrapper
+    return decorator

cartesia/version.py

@@ -1 +1 @@
-__version__ = "0.0.3"
+__version__ = "0.0.5"

cartesia-0.0.5.dist-info/METADATA

@@ -0,0 +1,187 @@
+Metadata-Version: 2.1
+Name: cartesia
+Version: 0.0.5
+Summary: The official Python library for the Cartesia API.
+Home-page: 
+Author: Cartesia, Inc.
+Author-email: support@cartesia.ai
+Classifier: Programming Language :: Python
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.8.0
+Description-Content-Type: text/markdown
+Requires-Dist: aiohttp
+Requires-Dist: httpx
+Requires-Dist: pytest-asyncio
+Requires-Dist: requests
+Requires-Dist: websockets
+Provides-Extra: all
+Requires-Dist: pytest >=8.0.2 ; extra == 'all'
+Requires-Dist: pytest-cov >=4.1.0 ; extra == 'all'
+Requires-Dist: twine ; extra == 'all'
+Requires-Dist: setuptools ; extra == 'all'
+Requires-Dist: wheel ; extra == 'all'
+Provides-Extra: dev
+Requires-Dist: pytest >=8.0.2 ; extra == 'dev'
+Requires-Dist: pytest-cov >=4.1.0 ; extra == 'dev'
+Requires-Dist: twine ; extra == 'dev'
+Requires-Dist: setuptools ; extra == 'dev'
+Requires-Dist: wheel ; extra == 'dev'
+
+
+# Cartesia Python API Library
+The official Cartesia Python library which provides convenient access to the Cartesia REST and Websocket API from any Python 3.8+ application.
+
+**Note:** This API is still in alpha. Please expect breaking changes and report any issues you encounter.
+
+## Installation
+```bash
+pip install cartesia
+
+# pip install in editable mode w/ dev dependencies
+pip install -e '.[dev]'
+```
+
+## Usage
+```python
+from cartesia.tts import CartesiaTTS
+import pyaudio
+import os
+
+client = CartesiaTTS(api_key=os.environ.get("CARTESIA_API_KEY"))
+voices = client.get_voices()
+voice = client.get_voice_embedding(voice_id=voices["Graham"]["id"])
+transcript = "Hello! Welcome to Cartesia"
+model_id = "genial-planet-1346" # (Optional) We'll specify a default if you don't have a specific model in mind
+
+p = pyaudio.PyAudio()
+
+stream = None
+
+# Generate and stream audio
+for output in client.generate(transcript=transcript, voice=voice, model_id=model_id, stream=True):
+    buffer = output["audio"]
+    rate = output["sampling_rate"]
+
+    if not stream:
+        stream = p.open(format=pyaudio.paFloat32,
+                        channels=1,
+                        rate=rate,
+                        output=True)
+
+    # Write the audio data to the stream
+    stream.write(buffer)
+
+stream.stop_stream()
+stream.close()
+p.terminate()
+```
+
+You can also use the async client if you want to make asynchronous API calls:
+```python
+from cartesia.tts import AsyncCartesiaTTS
+import asyncio
+import pyaudio
+import os
+
+async def write_stream():
+    client = AsyncCartesiaTTS(api_key=os.environ.get("CARTESIA_API_KEY"))
+    voices = client.get_voices()
+    voice = client.get_voice_embedding(voice_id=voices["Graham"]["id"])
+    transcript = "Hello! Welcome to Cartesia"
+    model_id = "genial-planet-1346" # (Optional) We'll specify a default if you don't have a specific model in mind
+
+    p = pyaudio.PyAudio()
+
+    stream = None
+
+    # Generate and stream audio
+    async for output in await client.generate(transcript=transcript, voice=voice, model_id=model_id, stream=True):
+        buffer = output["audio"]
+        rate = output["sampling_rate"]
+
+        if not stream:
+            stream = p.open(format=pyaudio.paFloat32,
+                            channels=1,
+                            rate=rate,
+                            output=True)
+
+        # Write the audio data to the stream
+        stream.write(buffer)
+
+    stream.stop_stream()
+    stream.close()
+    p.terminate()
+
+asyncio.run(write_stream())
+```
+
+If you are using Jupyter Notebook or JupyterLab, you can use IPython.display.Audio to play the generated audio directly in the notebook.
+Additionally, in these notebook examples we show how to use the client as a context manager (though this is not required).
+
+```python
+from IPython.display import Audio
+import io
+import os
+import numpy as np
+
+from cartesia.tts import CartesiaTTS
+
+with CartesiaTTS(api_key=os.environ.get("CARTESIA_API_KEY")) as client:
+    voices = client.get_voices()
+    voice = client.get_voice_embedding(voice_id=voices["Graham"]["id"])
+    transcript = "Hello! Welcome to Cartesia"
+
+    # Create a BytesIO object to store the audio data
+    audio_data = io.BytesIO()
+
+    # Generate and stream audio
+    for output in client.generate(transcript=transcript, voice=voice, stream=True):
+        buffer = output["audio"]
+        audio_data.write(buffer)
+
+# Set the cursor position to the beginning of the BytesIO object
+audio_data.seek(0)
+
+# Create an Audio object from the BytesIO data
+audio = Audio(np.frombuffer(audio_data.read(), dtype=np.float32), rate=output["sampling_rate"])
+
+# Display the Audio object
+display(audio)
+```
+
+Below is the same example using the async client:
+```python
+from IPython.display import Audio
+import io
+import os
+import numpy as np
+
+from cartesia.tts import AsyncCartesiaTTS
+
+async with AsyncCartesiaTTS(api_key=os.environ.get("CARTESIA_API_KEY")) as client:
+    voices = client.get_voices()
+    voice = client.get_voice_embedding(voice_id=voices["Graham"]["id"])
+    transcript = "Hello! Welcome to Cartesia"
+
+    # Create a BytesIO object to store the audio data
+    audio_data = io.BytesIO()
+
+    # Generate and stream audio
+    async for output in await client.generate(transcript=transcript, voice=voice, stream=True):
+        buffer = output["audio"]
+        audio_data.write(buffer)
+
+# Set the cursor position to the beginning of the BytesIO object
+audio_data.seek(0)
+
+# Create an Audio object from the BytesIO data
+audio = Audio(np.frombuffer(audio_data.read(), dtype=np.float32), rate=output["sampling_rate"])
+
+# Display the Audio object
+display(audio)
+```
+
+To avoid storing your API key in the source code, we recommend doing one of the following:
+1. Use [`python-dotenv`](https://pypi.org/project/python-dotenv/) to add `CARTESIA_API_KEY="my-api-key"` to your .env file. 
+1. Set the `CARTESIA_API_KEY` environment variable, preferably to a secure shell init file (e.g. `~/.zshrc`, `~/.bashrc`)

cartesia-0.0.5.dist-info/RECORD

@@ -0,0 +1,8 @@
+cartesia/__init__.py,sha256=uIc9xGNPs8_A6eAvbTUY1geazunYoEZVWFKhCwC9TRA,102
+cartesia/tts.py,sha256=Gtm9qse83g3SX4-KbmlxOAvTQcZIjmUkMfBKu2Xf9rY,26449
+cartesia/utils.py,sha256=GoTJe8LZ3WpS4hXkwoZauPYjo7Mbx7BvbBjAX5vEbwg,3024
+cartesia/version.py,sha256=S7u1lbuWmM3A3ajykBialmPoJUK6Jg-WmNqM-9OZFdk,22
+cartesia-0.0.5.dist-info/METADATA,sha256=oK64bcTyLhrosXh9FjuEwB2SUdQzbYsxzOWCnf6qaI4,5974
+cartesia-0.0.5.dist-info/WHEEL,sha256=DZajD4pwLWue70CAfc7YaxT1wLUciNBvN_TTcvXpltE,110
+cartesia-0.0.5.dist-info/top_level.txt,sha256=rTX4HnnCegMxl1FK9czpVC7GAvf3SwDzPG65qP-BS4w,9
+cartesia-0.0.5.dist-info/RECORD,,

{cartesia-0.0.3.dist-info → cartesia-0.0.5.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: bdist_wheel (0.41.2)
+Generator: bdist_wheel (0.43.0)
 Root-Is-Purelib: true
 Tag: py2-none-any
 Tag: py3-none-any

cartesia-0.0.3.dist-info/METADATA

@@ -1,113 +0,0 @@
-Metadata-Version: 2.1
-Name: cartesia
-Version: 0.0.3
-Summary: The official Python library for the Cartesia API.
-Home-page: 
-Author: Cartesia, Inc.
-Author-email: support@cartesia.ai
-Classifier: Programming Language :: Python
-Classifier: Programming Language :: Python :: 3
-Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
-Requires-Python: >=3.8.0
-Description-Content-Type: text/markdown
-Requires-Dist: websockets
-Requires-Dist: requests
-Provides-Extra: all
-Requires-Dist: pre-commit ; extra == 'all'
-Requires-Dist: docformatter ; extra == 'all'
-Requires-Dist: black ==24.1.1 ; extra == 'all'
-Requires-Dist: isort ==5.13.2 ; extra == 'all'
-Requires-Dist: flake8 ==7.0.0 ; extra == 'all'
-Requires-Dist: flake8-bugbear ==24.2.6 ; extra == 'all'
-Requires-Dist: pytest >=8.0.2 ; extra == 'all'
-Requires-Dist: pytest-cov >=4.1.0 ; extra == 'all'
-Provides-Extra: dev
-Requires-Dist: pre-commit ; extra == 'dev'
-Requires-Dist: docformatter ; extra == 'dev'
-Requires-Dist: black ==24.1.1 ; extra == 'dev'
-Requires-Dist: isort ==5.13.2 ; extra == 'dev'
-Requires-Dist: flake8 ==7.0.0 ; extra == 'dev'
-Requires-Dist: flake8-bugbear ==24.2.6 ; extra == 'dev'
-Requires-Dist: pytest >=8.0.2 ; extra == 'dev'
-Requires-Dist: pytest-cov >=4.1.0 ; extra == 'dev'
-
-
-# Cartesia Python API Library
-The official Cartesia Python library which provides convenient access to the Cartesia REST and Websocket API from any Python 3.8+ application.
-
-**Note:** This API is still in alpha. Please expect breaking changes and report any issues you encounter.
-
-## Installation
-```bash
-pip install cartesia
-
-# pip install in editable mode w/ dev dependencies
-pip install -e '.[dev]'
-```
-
-## Usage
-```python
-from cartesia.tts import CartesiaTTS
-import pyaudio
-import os
-
-client = CartesiaTTS(api_key=os.environ.get("CARTESIA_API_KEY"))
-voices = client.get_voices()
-voice = client.get_voice_embedding(voice_id=voices["Graham"]["id"])
-transcript = "Hello! Welcome to Cartesia"
-
-p = pyaudio.PyAudio()
-
-stream = None
-
-# Generate and stream audio
-for output in client.generate(transcript=transcript, voice=voice, stream=True):
-    buffer = output["audio"]
-    rate = output["sampling_rate"]
-
-    if not stream:
-        stream = p.open(format=pyaudio.paFloat32,
-                        channels=1,
-                        rate=rate,
-                        output=True)
-
-    # Write the audio data to the stream
-    stream.write(buffer)
-
-stream.stop_stream()
-stream.close()
-p.terminate()
-```
-
-If you are using Jupyter Notebook or JupyterLab, you can use IPython.display.Audio to play the generated audio directly in the notebook. Here's an example:
-
-```python
-from cartesia.tts import CartesiaTTS
-from IPython.display import Audio
-import io
-import os
-
-client = CartesiaTTS(api_key=os.environ.get("CARTESIA_API_KEY"))
-voices = client.get_voices()
-voice = client.get_voice_embedding(voice_id=voices["Graham"]["id"])
-transcript = "Hello! Welcome to Cartesia"
-
-# Create a BytesIO object to store the audio data
-audio_data = io.BytesIO()
-
-# Generate and stream audio
-for output in client.generate(transcript=transcript, voice=voice, stream=True):
-    buffer = output["audio"]
-    audio_data.write(buffer)
-
-# Set the cursor position to the beginning of the BytesIO object
-audio_data.seek(0)
-
-# Create an Audio object from the BytesIO data
-audio = Audio(audio_data, rate=output["sampling_rate"])
-
-# Display the Audio object
-display(audio)
-```
-
-We recommend using [`python-dotenv`](https://pypi.org/project/python-dotenv/) to add `CARTESIA_API_KEY="my-api-key"` to your .env file so that your API Key is not stored in the source code.

cartesia-0.0.3.dist-info/RECORD

@@ -1,7 +0,0 @@
-cartesia/__init__.py,sha256=m8BX-qLjsMoI_JZtgf3jNi8R3cBZqYy-z4oEhYeJLdI,64
-cartesia/tts.py,sha256=ABXW9rc8Pn0GTRvb_7DHZKMtbvhGUiqOgHmvztwlOnI,12033
-cartesia/version.py,sha256=4GZKi13lDTD25YBkGakhZyEQZWTER_OWQMNPoH_UM2c,22
-cartesia-0.0.3.dist-info/METADATA,sha256=VsCGL1sITbKqERihK2rzVm9WIY5EJ5nCS_CXQ0s14ns,3604
-cartesia-0.0.3.dist-info/WHEEL,sha256=iYlv5fX357PQyRT2o6tw1bN-YcKFFHKqB_LwHO5wP-g,110
-cartesia-0.0.3.dist-info/top_level.txt,sha256=rTX4HnnCegMxl1FK9czpVC7GAvf3SwDzPG65qP-BS4w,9
-cartesia-0.0.3.dist-info/RECORD,,