feat: add SILMA AI TTS plugin

livekit-agents/pyproject.toml

@@ -115,6 +115,7 @@ runway = ["livekit-plugins-runway>=1.8.0"]
 rtzr = ["livekit-plugins-rtzr>=1.8.0"]
 sarvam = ["livekit-plugins-sarvam>=1.8.0"]
 silero = ["livekit-plugins-silero>=1.8.0"]
+silma = ["livekit-plugins-silma>=1.8.0"]
 simli = ["livekit-plugins-simli>=1.8.0"]
 smallestai = ["livekit-plugins-smallestai>=1.8.0"]
 simplismart = ["livekit-plugins-simplismart>=1.8.0"]

livekit-plugins/livekit-plugins-silma/README.md

@@ -0,0 +1,105 @@
+# SILMA AI plugin for LiveKit Agents
+
+Support for Arabic and English speech synthesis with [SILMA AI](https://silma.ai/)
+TTS v2 — Modern Standard Arabic, the Saudi (Najdi) dialect, and English.
+
+More information is available in the docs for the
+[TTS integration](https://docs.livekit.io/agents/models/tts/plugins/silma/).
+
+## Installation
+
+```bash
+pip install livekit-plugins-silma
+```
+
+Or with the LiveKit Agents extra:
+
+```bash
+uv add "livekit-agents[silma]"
+```
+
+## Pre-requisites
+
+You'll need an API key from [SILMA](https://app.silma.ai/api-keys). Set it as an
+environment variable:
+
+```bash
+export SILMA_API_KEY="..."
+```
+
+## Usage
+
+```python
+from livekit.agents import AgentSession
+from livekit.plugins import silma
+
+session = AgentSession(
+    tts=silma.TTS(
+        model="silma-tts-v2-msa",
+        voice="sarah",
+    ),
+    # ... stt, llm, vad
+)
+```
+
+### Models and voices
+
+| Model | Language | Voices |
+| --- | --- | --- |
+| `silma-tts-v2-english` | English | `james`, `emma` |
+| `silma-tts-v2-msa` | Modern Standard Arabic | `sarah`, `salma`, `salwa`, `saja`, `sultan`, `salman`, `sulaiman`, `salim` |
+| `silma-tts-v2-ksa` | Arabic, Saudi (Najdi) dialect | same as MSA |
+
+### Cloned voices
+
+Upload a voice under **Custom Voices** at https://app.silma.ai/voices and pass
+its id along with your user id:
+
+```python
+silma.TTS(
+    model="silma-tts-v2-ksa",
+    voice="sarah",
+    user_id="...",
+    custom_audio_id="voice_1769817467123",
+)
+```
+
+### Pronunciation hints
+
+SILMA reads phone numbers, emails and links correctly when they are tagged in
+the text:
+
+```python
+await session.say(
+    "You can reach us on <STAG_PN>92005455</STAG_PN> or at <STAG_EMAIL>hi@silma.ai</STAG_EMAIL>."
+)
+```
+
+The plugin keeps these tags intact when it splits text, so a tag is never cut in
+half across two requests.
+
+Account-level pronunciation overrides configured at https://app.silma.ai/control
+are applied when you pass `user_id` and
+`enable_server_pronunciation_overrides=True`.
+
+## How it works
+
+`stream()` uses the realtime WebSocket API (`wss://api.silma.ai/tts/v2/ws/stream`)
+and sentence-tokenizes incoming LLM text so each request is a complete
+utterance. `synthesize()` uses the binary streaming endpoint
+(`POST /stream`).
+
+SILMA caps `text` at 250 characters per request, so longer input is split on
+word boundaries and sent as sequential requests concatenated into one audio
+segment.
+
+SILMA returns a 24 kHz mono float32 waveform; the plugin converts it to 16-bit
+PCM for the agent pipeline.
+
+Requests carry a `User-Agent` of
+`LiveKit-Agents-SILMA/<plugin version> livekit-agents/<framework version>` on
+both transports, so SILMA can tell plugin traffic apart from direct API use.
+
+WebSocket connections are pooled and reused across turns. If a pooled
+connection has gone stale, the plugin reconnects once and replays the utterance,
+which is safe because no audio has reached the caller at that point.

livekit-plugins/livekit-plugins-silma/livekit/plugins/silma/__init__.py

@@ -0,0 +1,69 @@
+# Copyright 2026 LiveKit, Inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""SILMA AI plugin for LiveKit Agents
+
+Arabic and English text-to-speech with SILMA TTS v2.
+
+See https://silma.ai/ for more information.
+"""
+
+from .models import (
+    ARABIC_VOICES,
+    DEFAULT_MODEL,
+    DEFAULT_VOICE,
+    ENGLISH_VOICES,
+    TTSArabicVoices,
+    TTSEnglishVoices,
+    TTSModels,
+    TTSVoices,
+)
+from .tts import TTS, ChunkedStream, SynthesizeStream
+from .version import __version__
+
+__all__ = [
+    "TTS",
+    "ChunkedStream",
+    "SynthesizeStream",
+    "TTSModels",
+    "TTSVoices",
+    "TTSArabicVoices",
+    "TTSEnglishVoices",
+    "ARABIC_VOICES",
+    "ENGLISH_VOICES",
+    "DEFAULT_MODEL",
+    "DEFAULT_VOICE",
+    "__version__",
+]
+
+from livekit.agents import Plugin
+
+from .log import logger
+
+
+class SilmaPlugin(Plugin):
+    def __init__(self) -> None:
+        super().__init__(__name__, __version__, __package__, logger)
+
+
+Plugin.register_plugin(SilmaPlugin())
+
+# Cleanup docs of unexported modules
+_module = dir()
+NOT_IN_ALL = [m for m in _module if m not in __all__]
+
+__pdoc__ = {}
+
+for n in NOT_IN_ALL:
+    __pdoc__[n] = False