cartesia 0.0.3__tar.gz → 0.0.4__tar.gz

{cartesia-0.0.3 → cartesia-0.0.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: cartesia
-Version: 0.0.3
+Version: 0.0.4
 Summary: The official Python library for the Cartesia API.
 Home-page: 
 Author: Cartesia, Inc.
@@ -110,4 +110,6 @@ audio = Audio(audio_data, rate=output["sampling_rate"])
 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.
+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-0.0.3 → cartesia-0.0.4}/README.md

@@ -76,4 +76,6 @@ audio = Audio(audio_data, rate=output["sampling_rate"])
 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.
+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-0.0.3 → cartesia-0.0.4}/cartesia/tts.py

@@ -31,7 +31,11 @@ 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.
+
+    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:
 
@@ -55,18 +59,22 @@ class CartesiaTTS:
         ...     audio, sr = audio_chunk["audio"], audio_chunk["sampling_rate"]
     """
 
-    def __init__(self, *, api_key: str = None):
+    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.experimental_ws_handle_interrupts = experimental_ws_handle_interrupts
         self.refresh_websocket()
 
     def get_voices(self, skip_embeddings: bool = True) -> Dict[str, VoiceMetadata]:
@@ -167,8 +175,11 @@ class CartesiaTTS:
         """
         if self.websocket and not self._is_websocket_closed():
             self.websocket.close()
+        route = "audio/websocket"
+        if self.experimental_ws_handle_interrupts:
+            route = f"experimental/{route}"
         self.websocket = connect(
-            f"{self._ws_url()}/audio/websocket?api_key={self.api_key}",
+            f"{self._ws_url()}/{route}?api_key={self.api_key}",
             close_timeout=None,
         )
 
@@ -280,21 +291,50 @@ 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"]:
+            while True:
+                response = json.loads(self.websocket.recv())
+                if response["done"]:
+                    break
                 audio = base64.b64decode(response["data"])
-                # print("timing", time.perf_counter() - start)
-                yield {"audio": audio, "sampling_rate": response["sampling_rate"]}
 
-                response = json.loads(self.websocket.recv())
-        except Exception:
-            raise RuntimeError(f"Failed to generate audio. {response}")
+                optional_kwargs = {}
+                if include_context_id:
+                    optional_kwargs["context_id"] = response["context_id"]
+
+                yield {
+                    "audio": audio,
+                    "sampling_rate": response["sampling_rate"],
+                    **optional_kwargs,
+                }
+
+                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:
+            raise RuntimeError(f"Failed to generate audio. {response}") from e
 
     def _http_url(self):
         prefix = "http" if "localhost" in self.base_url else "https"

cartesia-0.0.4/cartesia/version.py

@@ -0,0 +1 @@
+__version__ = "0.0.4"

{cartesia-0.0.3 → cartesia-0.0.4}/cartesia.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: cartesia
-Version: 0.0.3
+Version: 0.0.4
 Summary: The official Python library for the Cartesia API.
 Home-page: 
 Author: Cartesia, Inc.
@@ -110,4 +110,6 @@ audio = Audio(audio_data, rate=output["sampling_rate"])
 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.
+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-0.0.3 → cartesia-0.0.4}/setup.py

@@ -78,7 +78,8 @@ class UploadCommand(Command):
     """Support setup.py upload."""
 
     description = "Build and publish the package."
-    user_options = []
+    user_options = [("skip-upload", "u", "skip git tagging and pypi upload")]
+    boolean_options = ["skip-upload"]
 
     @staticmethod
     def status(s):
@@ -86,21 +87,26 @@ class UploadCommand(Command):
         print("\033[1m{0}\033[0m".format(s))
 
     def initialize_options(self):
-        pass
+        self.skip_upload = False
 
     def finalize_options(self):
-        pass
+        self.skip_upload = bool(self.skip_upload)
 
     def run(self):
         try:
             self.status("Removing previous builds…")
             rmtree(os.path.join(here, "dist"))
+            rmtree(os.path.join(here, "build"))
         except OSError:
             pass
 
         self.status("Building Source and Wheel (universal) distribution…")
         os.system("{0} setup.py sdist bdist_wheel --universal".format(sys.executable))
 
+        if self.skip_upload:
+            self.status("Skipping git tagging and pypi upload")
+            sys.exit()
+
         self.status("Uploading the package to PyPI via Twine…")
         os.system("twine upload dist/*")
 
@@ -116,6 +122,9 @@ class BumpVersionCommand(Command):
     To use: python setup.py bumpversion -v <version>
 
     This command will push the new version directly and tag it.
+
+    Usage:
+        python setup.py bumpversion --version=1.0.1
     """
 
     description = "Installs the foo."
@@ -130,6 +139,11 @@ class BumpVersionCommand(Command):
 
     def initialize_options(self):
         self.version = None
+        self.base_branch = None
+        self.version_branch = None
+        self.updated_files = [
+            "cartesia/version.py",
+        ]
 
     def finalize_options(self):
         # This package cannot be imported at top level because it
@@ -147,14 +161,18 @@ class BumpVersionCommand(Command):
             )
 
     def _undo(self):
-        os.system(f"git restore --staged {PACKAGE_DIR}/__init__.py")
-        os.system(f"git checkout -- {PACKAGE_DIR}/__init__.py")
+        os.system(f"git restore --staged {' '.join(self.updated_files)}")
+        os.system(f"git checkout -- {' '.join(self.updated_files)}")
+
+        # Return to the original branch
+        os.system(f"git checkout {self.base_branch}")
+        os.system(f"git branch -D {self.version_branch}")
 
     def run(self):
         current_version = about["__version__"]
 
         self.status("Checking current branch is 'main'")
-        current_branch = get_git_branch()
+        self.base_branch = current_branch = get_git_branch()
         if current_branch != "main":
             raise RuntimeError(
                 "You can only bump the version from the 'main' branch. "
@@ -174,18 +192,25 @@ class BumpVersionCommand(Command):
 
         # TODO: Add check to see if all tests are passing on main.
 
+        # Checkout new branch
+        self.version_branch = f"bumpversion/v{self.version}"
+        self.status(f"Create branch '{self.version_branch}'")
+        err_code = os.system(f"git checkout -b {self.version_branch}")
+        if err_code != 0:
+            raise RuntimeError("Failed to create branch.")
+
         # Change the version in __init__.py
         self.status(f"Updating version {current_version} -> {self.version}")
         update_version(self.version)
-        if current_version != self.version:
-            self._undo()
-            raise RuntimeError("Failed to update version.")
+        # if current_version != self.version:
+        #     self._undo()
+        #     raise RuntimeError("Failed to update version.")
 
-        self.status(f"Adding {PACKAGE_DIR}/__init__.py to git")
-        err_code = os.system(f"git add {PACKAGE_DIR}/__init__.py")
+        self.status(f"Adding {', '.join(self.updated_files)} to git")
+        err_code = os.system(f"git add {' '.join(self.updated_files)}")
         if err_code != 0:
             self._undo()
-            raise RuntimeError("Failed to add file to git.")
+            raise RuntimeError("Failed to add files to git.")
 
         # Commit the file with a message '[bumpversion] v<version>'.
         self.status(f"Commit with message '[bumpversion] v{self.version}'")
@@ -195,12 +220,15 @@ class BumpVersionCommand(Command):
             raise RuntimeError("Failed to commit file to git.")
 
         # Push the commit to origin.
-        # self.status("Pushing commit to origin")
-        # err_code = os.system("git push")
-        # if err_code != 0:
-        #     # TODO: undo the commit automatically.
-        #     raise RuntimeError("Failed to push commit to origin.")
+        self.status(f"Pushing commit to origin/{self.version_branch}")
+        err_code = os.system(f"git push --force --set-upstream origin {self.version_branch}")
+        if err_code != 0:
+            # TODO: undo the commit automatically.
+            self._undo()
+            raise RuntimeError("Failed to push commit to origin.")
 
+        os.system(f"git checkout {self.base_branch}")
+        os.system(f"git branch -D {self.version_branch}")
         sys.exit()
 
 

{cartesia-0.0.3 → cartesia-0.0.4}/tests/test_tts.py

@@ -6,11 +6,12 @@ but rather for general correctness.
 """
 
 import os
-from typing import Dict, Generator
+import uuid
+from typing import Dict, Generator, List
 
 import pytest
 
-from cartesia.tts import CartesiaTTS, VoiceMetadata
+from cartesia.tts import DEFAULT_MODEL_ID, CartesiaTTS, VoiceMetadata
 
 SAMPLE_VOICE = "Milo"
 
@@ -26,6 +27,13 @@ def client():
     return CartesiaTTS(api_key=os.environ.get("CARTESIA_API_KEY"))
 
 
+@pytest.fixture(scope="session")
+def client_with_ws_interrupt():
+    return CartesiaTTS(
+        api_key=os.environ.get("CARTESIA_API_KEY"), experimental_ws_handle_interrupts=True
+    )
+
+
 @pytest.fixture(scope="session")
 def resources(client: CartesiaTTS):
     voices = client.get_voices()
@@ -93,6 +101,45 @@ def test_generate_stream(resources: _Resources, websocket: bool):
         assert isinstance(output["sampling_rate"], int)
 
 
+@pytest.mark.parametrize(
+    "actions",
+    [
+        ["cancel-5", None],
+        ["cancel-5", "cancel-1", None],
+        [None, "cancel-3", None],
+        [None, "cancel-1", "cancel-2"],
+    ],
+)
+def test_generate_stream_interrupt(
+    client_with_ws_interrupt: CartesiaTTS, resources: _Resources, actions: List[str]
+):
+    client = client_with_ws_interrupt
+    voices = resources.voices
+    embedding = voices[SAMPLE_VOICE]["embedding"]
+    transcript = "Hello, world!"
+
+    context_ids = [f"test-{uuid.uuid4().hex[:6]}" for _ in range(len(actions))]
+
+    for context_id, action in zip(context_ids, actions):
+        body = dict(transcript=transcript, model_id=DEFAULT_MODEL_ID, voice=embedding)
+
+        # Parse actions to see what we should expect.
+        if action is None:
+            num_turns = None
+        elif "cancel" in action:
+            num_turns = int(action.split("-")[1])
+
+        generator = client._generate_ws(body, context_id=context_id)
+        for idx, response in enumerate(generator):
+            assert response.keys() == {"audio", "sampling_rate", "context_id"}
+            assert response["context_id"] == context_id, (
+                f"Context ID from response ({response['context_id']}) does not match "
+                f"the expected context ID ({context_id})"
+            )
+            if idx + 1 == num_turns:
+                break
+
+
 @pytest.mark.parametrize("chunk_time", [0.05, 0.6])
 def test_check_inputs_invalid_chunk_time(client: CartesiaTTS, chunk_time):
     with pytest.raises(ValueError, match="`chunk_time` must be between 0.1 and 0.5"):

cartesia-0.0.3/cartesia/version.py

@@ -1 +0,0 @@
-__version__ = "0.0.3"