Harden NVIDIA Speech STT and TTS support

livekit-agents/livekit/agents/stt/stt.py

@@ -1,3 +1,6 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES.
+# SPDX-License-Identifier: Apache-2.0
+
 from __future__ import annotations
 
 import asyncio
@@ -235,7 +238,7 @@ async def recognize(
 
             except APIError as e:
                 retry_interval = conn_options._interval_for_retry(i)
-                if conn_options.max_retry == 0:
+                if conn_options.max_retry == 0 or not e.retryable:
                     self._emit_error(e, recoverable=False)
                     raise
                 elif i == conn_options.max_retry:
@@ -465,7 +468,7 @@ async def _main_task(self) -> None:
                 last_start_time = time.time()
                 return await self._run()
             except APIError as e:
-                if max_retries == 0:
+                if max_retries == 0 or not e.retryable:
                     self._emit_error(e, recoverable=False)
                     raise
                 elif self._num_retries == max_retries:

livekit-plugins/livekit-plugins-nvidia/README.md

@@ -1,8 +1,9 @@
-# NVIDIA plugin for LiveKit Agents
+<!-- SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. -->
+<!-- SPDX-License-Identifier: Apache-2.0 -->
 
-Support for [NVIDIA Riva](https://developer.nvidia.com/riva)'s speech AI services in LiveKit Agents.
+# NVIDIA plugin for LiveKit Agents
 
-More information is available in the [NVIDIA Riva documentation](https://developer.nvidia.com/riva).
+Support for NVIDIA Speech AI services in LiveKit Agents.
 
 ## Installation
 
@@ -15,4 +16,31 @@ pip install livekit-plugins-nvidia
 You can either:
 
 1. Use an API key from NVIDIA. It can be set as an environment variable: `NVIDIA_API_KEY`
-2. Use you self hosted [Nim](https://developer.nvidia.com/nim) server.
+2. Use your self-hosted [NIM](https://developer.nvidia.com/nim) server.
+
+## Usage
+
+```python
+from livekit.plugins import nvidia
+
+stt = nvidia.STT(
+    model="parakeet-1.1b-en-US-asr-streaming-silero-vad-sortformer",
+    inference_mode="streaming",
+    server="grpc.nvcf.nvidia.com:443",
+    use_ssl=True,
+    endpointing=nvidia.EndpointingConfig(mode="low_latency"),
+)
+
+tts = nvidia.TTS(
+    voice="Magpie-Multilingual.EN-US.Leo",
+    sample_rate=16000,
+    inference_mode="online",
+)
+```
+
+STT defaults to `inference_mode="auto"`; use `"streaming"` or `"offline"` to
+match the deployed model. TTS defaults to `"online"`; use `"offline"` for batch
+models.
+
+For local NVIDIA Speech deployments, pass the local `server` and set
+`use_ssl=False` when TLS is not enabled.

livekit-plugins/livekit-plugins-nvidia/livekit/plugins/nvidia/__init__.py

@@ -1,3 +1,6 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES.
+# SPDX-License-Identifier: Apache-2.0
+
 # Copyright 2025 LiveKit, Inc.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -35,11 +38,19 @@ def __getattr__(name: str) -> typing.Any:
     raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
 
 
-from .stt import STT, SpeechStream  # noqa: E402
+from .stt import STT, EndpointingConfig, SpeechStream  # noqa: E402
 from .tts import TTS, SynthesizeStream  # noqa: E402
 from .version import __version__  # noqa: E402
 
-__all__ = ["STT", "SpeechStream", "TTS", "SynthesizeStream", "realtime", "__version__"]
+__all__ = [
+    "STT",
+    "EndpointingConfig",
+    "SpeechStream",
+    "TTS",
+    "SynthesizeStream",
+    "realtime",
+    "__version__",
+]
 
 
 from livekit.agents import Plugin  # noqa: E402