RT SDK bugfixes and optimisations

.github/workflows/test.yaml

@@ -22,7 +22,7 @@ jobs:
         with:
           python-version: ${{ matrix.python-version }}
       - name: Install dependencies
-        run: make install-dev
+        run: make install-dev-rt
       - name: Lint RT SDK
         run: make lint-rt
       - name: Test RT SDK
@@ -42,7 +42,7 @@ jobs:
         with:
           python-version: ${{ matrix.python-version }}
       - name: Install dependencies
-        run: make install-dev
+        run: make install-dev-batch
       - name: Lint Batch SDK
         run: make lint-batch
       - name: Test Batch SDK
@@ -62,7 +62,7 @@ jobs:
         with:
           python-version: ${{ matrix.python-version }}
       - name: Install dependencies
-        run: make install-dev
+        run: make install-dev-flow
       - name: Lint Flow SDK
         run: make lint-flow
       - name: Test Flow SDK
@@ -82,7 +82,7 @@ jobs:
         with:
           python-version: ${{ matrix.python-version }}
       - name: Install dependencies
-        run: make install-dev
+        run: make install-dev-voice
       - name: Lint Voice Agent SDK
         run: make lint-voice
       - name: Test Voice Agent SDK
@@ -102,7 +102,7 @@ jobs:
         with:
           python-version: ${{ matrix.python-version }}
       - name: Install dependencies
-        run: make install-dev
+        run: make install-dev-tts
       - name: Lint TTS SDK
         run: make lint-tts
       - name: Test TTS SDK

Makefile

@@ -7,6 +7,7 @@
 .PHONY: type-check-all type-check-rt type-check-batch type-check-flow type-check-tts type-check-voice
 .PHONY: build-all build-rt build-batch build-flow build-tts build-voice
 .PHONY: clean-all clean-rt clean-batch clean-flow clean-tts clean-voice
+.PHONY: install-dev install-dev-rt install-dev-batch install-dev-flow install-dev-tts install-dev-voice
 
 
 help:
@@ -145,13 +146,33 @@ type-check-voice:
 	cd sdk/voice/speechmatics && mypy .
 
 # Installation targets
-install-dev:
+#
+# voice depends on speechmatics-rt>=0.5.3; the local rt package's own dev version (0.0.0)
+# never satisfies that, so pip silently replaces the editable rt install with a published
+# PyPI wheel the moment voice[dev] installs after it. rt must always be the LAST install
+# in any sequence that also installs voice, so the local editable copy is what's left active.
+install-dev: install-dev-batch install-dev-flow install-dev-tts install-dev-voice install-dev-rt
+
+install-dev-rt:
 	python -m pip install --upgrade pip
 	python -m pip install -e sdk/rt[dev]
+
+install-dev-batch:
+	python -m pip install --upgrade pip
 	python -m pip install -e sdk/batch[dev]
+
+install-dev-flow:
+	python -m pip install --upgrade pip
 	python -m pip install -e sdk/flow[dev]
+
+install-dev-tts:
+	python -m pip install --upgrade pip
 	python -m pip install -e sdk/tts[dev]
+
+install-dev-voice:
+	python -m pip install --upgrade pip
 	python -m pip install -e sdk/voice[dev]
+	python -m pip install -e sdk/rt
 
 install-build:
 	python -m pip install --upgrade build

sdk/rt/speechmatics/rt/_async_client.py

@@ -10,7 +10,6 @@
 from ._base_client import _BaseClient
 from ._exceptions import AudioError
 from ._exceptions import TimeoutError
-from ._exceptions import TranscriptionError
 from ._logging import get_logger
 from ._models import AudioEncoding
 from ._models import AudioEventsConfig
@@ -39,6 +38,9 @@ class AsyncClient(_BaseClient):
         url: WebSocket endpoint URL. If not provided, uses SPEECHMATICS_RT_URL
                 environment variable or defaults to EU endpoint.
         conn_config: Websocket connection configuration.
+        sdk_identifier: Value reported to the service as `sm-sdk`. For a package built on
+                top of speechmatics-rt, pass its own identifier here instead of subclassing
+                Transport to override `_prepare_url`.
 
     Raises:
         ConfigurationError: If required configuration is missing or invalid.
@@ -76,22 +78,26 @@ def __init__(
         api_key: Optional[str] = None,
         url: Optional[str] = None,
         conn_config: Optional[ConnectionConfig] = None,
+        sdk_identifier: Optional[str] = None,
     ) -> None:
         self._logger = get_logger("speechmatics.rt.async_client")
 
+        self._build_transport = lambda request_id: self._create_transport_from_config(
+            auth=auth,
+            api_key=api_key,
+            url=url,
+            conn_config=conn_config,
+            request_id=request_id,
+            sdk_identifier=sdk_identifier,
+        )
+
         (
             self._session,
             self._recognition_started_evt,
             self._session_done_evt,
         ) = self._init_session_info()
 
-        transport = self._create_transport_from_config(
-            auth=auth,
-            api_key=api_key,
-            url=url,
-            conn_config=conn_config,
-            request_id=self._session.request_id,
-        )
+        transport = self._build_transport(self._session.request_id)
         super().__init__(transport)
 
         self.on(ServerMessageType.RECOGNITION_STARTED, self._on_recognition_started)
@@ -340,7 +346,12 @@ async def _send_eos(self, seq_no: int) -> None:
 
     async def _wait_recognition_started(self, timeout: float = 5.0) -> None:
         """Wait for RecognitionStarted message from server."""
-        await asyncio.wait_for(self._recognition_started_evt.wait(), timeout)
+        await self._wait_started_or_session_done(self._recognition_started_evt, timeout)
+
+    def _reset_session_events(self) -> None:
+        """Clear the session-tracking events, for a fresh session."""
+        self._recognition_started_evt.clear()
+        self._session_done_evt.clear()
 
     def _on_recognition_started(self, msg: dict[str, Any]) -> None:
         """Handle RecognitionStarted message from server."""
@@ -357,8 +368,8 @@ def _on_error(self, msg: dict[str, Any]) -> None:
         """Handle Error message from server."""
         error = msg.get("reason", "unknown")
         self._logger.error("Server error: %s", error)
+        self._last_error_reason = error
         self._session_done_evt.set()
-        raise TranscriptionError(error)
 
     def _on_audio_added(self, msg: dict[str, Any]) -> None:
         """Handle AudioAdded message from server."""

sdk/rt/speechmatics/rt/_async_multi_channel_client.py

@@ -10,7 +10,6 @@
 from ._base_client import _BaseClient
 from ._exceptions import ConfigurationError
 from ._exceptions import TimeoutError
-from ._exceptions import TranscriptionError
 from ._logging import get_logger
 from ._models import AudioEventsConfig
 from ._models import AudioFormat
@@ -41,6 +40,8 @@ class AsyncMultiChannelClient(_BaseClient):
         url: WebSocket endpoint URL. If not provided, uses SPEECHMATICS_RT_URL
                 environment variable or defaults to EU endpoint.
         conn_config: Websocket connection configuration.
+        sdk_identifier: Value reported to the service as `sm-sdk`, for a package built on
+                top of speechmatics-rt.
 
     Examples:
         Transcribing stereo audio:
@@ -65,23 +66,27 @@ def __init__(
         api_key: Optional[str] = None,
         url: Optional[str] = None,
         conn_config: Optional[ConnectionConfig] = None,
+        sdk_identifier: Optional[str] = None,
     ) -> None:
         self._logger = get_logger("speechmatics.rt.async_multi_chan_client")
 
+        self._build_transport = lambda request_id: self._create_transport_from_config(
+            auth=auth,
+            api_key=api_key,
+            url=url,
+            conn_config=conn_config,
+            request_id=request_id,
+            sdk_identifier=sdk_identifier,
+        )
+
         (
             self._session,
             self._rec_started_evt,
             self._session_done_evt,
         ) = self._init_session_info()
         self._eos_sent = False
 
-        transport = self._create_transport_from_config(
-            auth=auth,
-            api_key=api_key,
-            url=url,
-            conn_config=conn_config,
-            request_id=self._session.request_id,
-        )
+        transport = self._build_transport(self._session.request_id)
 
         super().__init__(transport)
 
@@ -182,7 +187,12 @@ async def transcribe(
 
     async def _wait_recognition_started(self, timeout: float = 5.0) -> None:
         """Wait for RecognitionStarted message from server."""
-        await asyncio.wait_for(self._rec_started_evt.wait(), timeout=timeout)
+        await self._wait_started_or_session_done(self._rec_started_evt, timeout)
+
+    def _reset_session_events(self) -> None:
+        """Clear the session-tracking events, for a fresh session."""
+        self._rec_started_evt.clear()
+        self._session_done_evt.clear()
 
     async def _audio_producer(self, sources: dict[str, BinaryIO], chunk_size: int) -> None:
         """
@@ -247,8 +257,10 @@ def _on_eot(self, msg: dict[str, Any]) -> None:
 
     def _on_error(self, msg: dict[str, Any]) -> None:
         """Handle Error message from server."""
+        error = msg.get("reason", "unknown")
+        self._logger.error("Server error: %s", error)
+        self._last_error_reason = error
         self._session_done_evt.set()
-        raise TranscriptionError(msg.get("reason", "unknown"))
 
     def _on_warning(self, msg: dict[str, Any]) -> None:
         """Handle Warning message from server."""

sdk/rt/speechmatics/rt/_auth.py

@@ -4,6 +4,7 @@
 from typing import Optional
 
 from ._exceptions import AuthenticationError
+from .constants import REQUEST_ID_HEADER
 
 
 class AuthBase(abc.ABC):
@@ -125,7 +126,7 @@ async def _generate_token(self) -> str:
         }
 
         if self._request_id:
-            headers["X-Request-Id"] = self._request_id
+            headers.setdefault(REQUEST_ID_HEADER, self._request_id)
 
         try:
             async with aiohttp.ClientSession() as session: