sayna-client 0.0.18

Python SDK for Sayna server-side WebSocket connections

Sayna Python SDK

Python SDK for Sayna's real-time voice interaction API. Send audio for speech recognition, receive synthesized speech, and manage voice sessions from your Python applications.

Features

• 🎤 Speech-to-Text: Real-time transcription with support for multiple providers (Deepgram, Google, etc.)
• 🔊 Text-to-Speech: High-quality voice synthesis with various TTS providers (ElevenLabs, Google, etc.)
• 🔌 WebSocket Connection: Async/await support with aiohttp
• 🌐 REST API: Standalone endpoints for health checks, voice catalog, TTS synthesis, and SIP hooks management
• 🔐 Webhook Receiver: Secure verification and parsing of SIP webhooks with HMAC-SHA256 signatures
• ✅ Type Safety: Full type hints with Pydantic models
• 🚀 Easy to Use: Simple, intuitive API
• 📦 Modern Python: Built for Python 3.9+

Installation

pip install sayna-client

Room Scoping

Room names are sent to the server as-is; the SDK does not modify or prefix them. Access scoping is enforced server-side based on room metadata:

• Room listings (get_livekit_rooms) are scoped to the authenticated context and may return fewer rooms than exist on the server.
• Room operations (get_livekit_room, remove_livekit_participant, mute_livekit_participant_track, sip_transfer_rest) will return 404 if the room is not found or not accessible to the current authentication context.
• Token generation (get_livekit_token) will return 403 if attempting to create a token for a room owned by a different tenant.
• Inbound SIP-created rooms are owned by the routing configuration's auth_id and may not be accessible under other authentication contexts.

Do not attempt to work around access errors by modifying room names—the SDK does not rewrite them and the server enforces ownership.

Quick Start

import asyncio
from sayna_client import SaynaClient, STTConfig, TTSConfig

async def main():
    # Initialize the client with configs
    client = SaynaClient(
        url="https://api.sayna.ai",
        stt_config=STTConfig(
            provider="deepgram",
            model="nova-2"
        ),
        tts_config=TTSConfig(
            provider="cartesia",
            voice_id="example-voice"
        ),
        api_key="your-api-key"
    )

    # Register callbacks
    client.register_on_stt_result(lambda result: print(f"Transcription: {result.transcript}"))
    client.register_on_tts_audio(lambda audio: print(f"Received {len(audio)} bytes of audio"))

    # Connect and interact
    await client.connect()
    await client.speak("Hello, world!")
    await client.disconnect()

if __name__ == "__main__":
    asyncio.run(main())

The constructor also accepts an optional loading_audio=LoadingAudioConfig(...) keyword argument for a server-side "thinking" audio loop on a dedicated LiveKit track; see Loading Indicator below.

API

REST API Methods

These methods use HTTP endpoints and don't require an active WebSocket connection:

await client.health()

Performs a health check on the Sayna server.

Returns: HealthResponse - Response object with status field ("OK" when healthy).

Example:

health = await client.health()
print(health.status)  # "OK"

---

await client.get_voices()

Retrieves the catalogue of text-to-speech voices grouped by provider.

Returns: dict[str, list[VoiceDescriptor]] - Dictionary where keys are provider names and values are lists of voice descriptors.

Example:

voices = await client.get_voices()
for provider, voice_list in voices.items():
    print(f"{provider}:", [v.name for v in voice_list])

---

await client.speak_rest(text, tts_config)

Synthesizes text into audio using the REST API. This is a standalone method that doesn't require an active WebSocket connection.

Parameter	Type	Purpose
text	str	Text to synthesize (must be non-empty).
tts_config	TTSConfig	Text-to-speech provider configuration.

Returns: tuple[bytes, dict[str, str]] - Tuple of (audio_data, response_headers). Headers include: Content-Type, Content-Length, x-audio-format, x-sample-rate.

Example:

audio_data, headers = await client.speak_rest("Hello, world!", TTSConfig(
    provider="elevenlabs",
    voice_id="21m00Tcm4TlvDq8ikWAM",
    model="eleven_turbo_v2",
    speaking_rate=1.0,
    audio_format="mp3",
    sample_rate=24000,
    connection_timeout=30,
    request_timeout=60,
    pronunciations=[]
))
print(f"Received {len(audio_data)} bytes of {headers['Content-Type']}")

---

await client.get_livekit_token(room_name, participant_name, participant_identity)

Issues a LiveKit access token for a participant. Room names are sent as-is; the SDK does not modify them. Access scoping is enforced server-side.

Parameter	Type	Purpose
room_name	str	LiveKit room to join or create.
participant_name	str	Display name for the participant.
participant_identity	str	Unique identifier for the participant.

Returns: LiveKitTokenResponse - Object containing token, room name, participant identity, and LiveKit URL.

Raises: SaynaHttpError with status 403 if the room exists but belongs to a different authenticated context (ownership conflict). Do not retry with modified names.

Example:

token_info = await client.get_livekit_token(
    "my-room",
    "John Doe",
    "user-123"
)
print("Token:", token_info.token)
print("LiveKit URL:", token_info.livekit_url)

---

await client.get_sip_hooks()

Retrieves all configured SIP webhook hooks from the runtime cache.

Returns: SipHooksResponse - Object containing the list of configured SIP hooks.

Example:

hooks = await client.get_sip_hooks()
for hook in hooks.hooks:
    print(f"{hook.host} -> {hook.url} (auth_id: {hook.auth_id})")

---

await client.set_sip_hooks(hooks)

Adds or replaces SIP webhook hooks. Existing hooks with matching hosts (case-insensitive) are replaced.

Parameter	Type	Purpose
hooks	list[SipHook]	List of SipHook objects to add or replace.

Each SipHook requires an auth_id field that associates inbound SIP calls with a tenant for room ownership. When AUTH_REQUIRED=true on the server, auth_id must be non-empty. When AUTH_REQUIRED=false, auth_id may be empty but must still be provided.

Returns: SipHooksResponse - Object containing the merged list of all hooks (existing + new).

Example:

from sayna_client import SipHook

hooks = [
    SipHook(
        host="example.com",
        url="https://webhook.example.com/events",
        auth_id="tenant-123",
    ),
    SipHook(
        host="another.com",
        url="https://webhook.another.com/events",
        auth_id="tenant-456",
    ),
]
response = await client.set_sip_hooks(hooks)
print(f"Total hooks: {len(response.hooks)}")

---

WebSocket API Methods

These methods require an active WebSocket connection:

SaynaClient(url, stt_config, tts_config, livekit_config=None, without_audio=False, api_key=None, stream_id=None, loading_audio=None)

Creates a new SaynaClient instance.

Parameter	Type	Default	Purpose
url	str	-	Sayna server URL (http://, https://, ws://, or wss://).
stt_config	STTConfig	-	Speech-to-text provider configuration.
tts_config	TTSConfig	-	Text-to-speech provider configuration.
livekit_config	LiveKitConfig (optional)	None	Optional LiveKit room configuration.
without_audio	bool	False	Disable audio streaming.
api_key	str (optional)	None	API key for authentication.
stream_id	str (optional)	None	Session identifier for recording paths; server generates a UUID when omitted.
loading_audio	LoadingAudioConfig (optional)	None	Server-side "thinking" audio clip looped on a dedicated LiveKit track when loading_start() is called. See Loading Indicator below.

---

await client.connect()

Establishes the WebSocket connection and sends the initial configuration. connect() does not wait for the ready event; check client.ready or register client.register_on_ready(callback) if you need to gate work on readiness.

---

client.register_on_stt_result(callback)

Registers a callback for speech-to-text transcription results.

Example:

def handle_stt(result):
    print(f"Transcription: {result.transcript}")

client.register_on_stt_result(handle_stt)

---

client.register_on_tts_audio(callback)

Registers a callback for text-to-speech audio data.

Example:

def handle_audio(audio_data: bytes):
    print(f"Received {len(audio_data)} bytes of audio")

client.register_on_tts_audio(handle_audio)

---

client.register_on_error(callback)

Registers a callback for error messages. Ownership and access errors from WebSocket operations may arrive as error callbacks. If you receive access errors, verify that you are using the correct room name and that the room belongs to your authenticated context. Do not retry with modified names—the SDK does not rewrite room names.

---

client.register_on_message(callback)

Registers a callback for participant messages.

---

client.register_on_participant_connected(callback)

Registers a callback for participant connection events.

---

client.register_on_participant_disconnected(callback)

Registers a callback for participant disconnection events.

---

client.register_on_track_subscribed(callback)

Registers a callback for track subscription events.

---

client.register_on_tts_playback_complete(callback)

Registers a callback for TTS playback completion events.

Unknown or malformed websocket control messages are logged and ignored so protocol drift does not break the receive loop. Actual server error messages still reach client.register_on_error(callback).

---

await client.speak(text, flush=True, allow_interruption=True)

Sends text to be synthesized as speech.

Parameter	Type	Default	Purpose
text	str	-	Text to synthesize.
flush	bool	True	Clear TTS queue before speaking.
allow_interruption	bool	True	Allow speech to be interrupted.

Example:

await client.speak("Hello, world!")
await client.speak("Important message", flush=True, allow_interruption=False)

---

await client.on_audio_input(audio_data)

Sends raw audio data (bytes) to the server for speech recognition.

Example:

await client.on_audio_input(audio_bytes)

---

await client.send_message(message, role, topic=None, debug=None)

Sends a message to the Sayna session with role and optional metadata.

Parameter	Type	Default	Purpose
message	str	-	Message content.
role	str	-	Sender role (e.g., 'user', 'assistant').
topic	str (optional)	None	LiveKit topic/channel. Omitted when not provided.
debug	dict (optional)	None	Optional debug metadata.

---

await client.clear()

Clears the text-to-speech queue.

---

await client.tts_flush(allow_interruption=True)

Flushes the TTS queue by sending an empty speak command.

---

await client.loading_start()

Tells the server to begin the seamless playback loop of the configured loading-audio clip on the dedicated "loading-audio" LiveKit track. Fire-and-forget: the call returns as soon as the WebSocket frame is sent, without waiting for a server acknowledgement.

Server-side this is idempotent: calling loading_start() while the loop is already running (including during the brief fade-out window of a prior loading_stop()) is a no-op.

Requires audio to be enabled, a LiveKit room to be configured, and a loading_audio argument supplied at construction time. The SDK does not pre-check these prerequisites; the server enforces them and reports failures through the error channel — see Loading Indicator.

Raises: SaynaNotConnectedError if not connected; SaynaNotReadyError if the session is not ready; SaynaConnectionError if the transport-layer send fails.

---

await client.loading_stop()

Tells the server to stop the loading-indicator audio loop with a short server-side fade-out. Fire-and-forget and always silent on the server side: the server never emits an error for loading_stop, even when no loop is running or no LiveKit room is configured.

Call this immediately before speak() if you do not want the loop to overlap with synthesized speech. See Loading Indicator for the full call flow.

Raises: SaynaNotConnectedError if not connected; SaynaNotReadyError if the session is not ready; SaynaConnectionError if the transport-layer send fails.

---

await client.disconnect()

Disconnects from the WebSocket server and cleans up resources.

---

Properties

• client.ready: Boolean indicating whether the connection is ready (received ready message).
• client.connected: Boolean indicating whether the WebSocket is connected.
• client.livekit_room_name: LiveKit room name (available after ready when LiveKit is enabled).
• client.livekit_url: LiveKit URL (available after ready).
• client.sayna_participant_identity: Sayna participant identity (available after ready when LiveKit is enabled).
• client.sayna_participant_name: Sayna participant name (available after ready when LiveKit is enabled).

---

Loading Indicator

The Sayna server can play a short audio clip as a seamless loop while your application is busy ("thinking"), giving the caller an audible equivalent of a spinner. The loop is published on a dedicated LiveKit audio track named "loading-audio" that is independent of the speech track ("tts-audio"), so it never interferes with STT or TTS. The SDK does not decode, parse, or play the audio itself; it forwards the configuration to the server, which performs all decoding, validation, volume scaling, and looping.

The authoritative protocol contract lives in ../sayna/docs/websocket.md#loading-indicator; see also ../sayna/docs/api-reference.md and the server change in SaynaAI/sayna#18.

LoadingAudioConfig

Pydantic model passed once to the constructor via loading_audio=. Unknown fields are rejected (extra="forbid").

from typing import Literal, Optional
from pydantic import BaseModel

class LoadingAudioConfig(BaseModel):
    data: str                                   # required, base64-encoded
    format: Optional[Literal["wav", "pcm"]] = None
    sample_rate: Optional[int] = None
    channels: Optional[int] = None
    volume: Optional[float] = None

Field	Type	Purpose
data	str (required)	Base64-encoded audio bytes (standard alphabet, padded): either a complete WAV file or raw 16-bit little-endian PCM.
format	Literal["wav", "pcm"] (optional)	Container hint. Omit to let the server auto-detect from the RIFF/WAVE signature. Any other value is rejected by the server.
sample_rate	int (optional)	Sample rate in Hz (8000-48000). Required for raw PCM (no header is present); ignored for WAV (the header is authoritative).
channels	int (optional)	Channel count for raw PCM (typically 1 or 2). Defaults to 1 server-side. Ignored for WAV.
volume	float (optional)	Playback volume in [0.0, 1.0]. Defaults to 1.0. Out-of-range values are clamped by the server (not rejected). Applied once at config time as amplitude scaling; there is no runtime volume control.

The server is authoritative on every audio-content rule (duration, bit depth, byte cap, channel count, sample-rate range). The SDK does not mirror those limits to avoid drift.

Loading a WAV file and base64-encoding it

The SDK ships no file-loading helper; encoding lives in your application code:

import base64

with open("loading.wav", "rb") as f:
    data = base64.b64encode(f.read()).decode("ascii")

Full call flow

import asyncio
import base64
from sayna_client import (
    SaynaClient,
    STTConfig,
    TTSConfig,
    LiveKitConfig,
    LoadingAudioConfig,
)

async def main():
    with open("loading.wav", "rb") as f:
        loading_data = base64.b64encode(f.read()).decode("ascii")

    client = SaynaClient(
        url="https://api.sayna.ai",
        stt_config=STTConfig(provider="deepgram", model="nova-2"),
        tts_config=TTSConfig(provider="cartesia", voice_id="example-voice"),
        livekit_config=LiveKitConfig(room_name="my-room"),
        loading_audio=LoadingAudioConfig(data=loading_data, format="wav"),
        api_key="your-api-key",
    )

    client.register_on_error(lambda err: print(f"Server error: {err.message}"))

    await client.connect()

    # User turn complete: start the loop while the application thinks.
    await client.loading_start()

    # ...application does its background work (LLM call, database lookup, etc.)...

    # IMPORTANT: stop the loop BEFORE calling speak(). The server and SDK
    # deliberately do not auto-stop the loop on speak()/clear() so callers
    # control the overlap explicitly. Skipping loading_stop() will play the
    # loop and the spoken answer simultaneously on separate LiveKit tracks.
    await client.loading_stop()
    await client.speak("Here is the answer.")

    await client.disconnect()

asyncio.run(main())

Error channel

Loading-indicator failures arrive on the existing register_on_error(callback) channel as plain ErrorMessage instances; there is no separate loading_error event, no register_on_loading_error callback, and no dedicated exception class. The following conditions all surface this way:

• A LoadingAudioConfig was supplied but the server failed to decode it (invalid base64, unsupported format, sample rate or channel count out of range, clip too long, byte cap exceeded). The decode happens once at config time, and the session stays alive afterwards.
• loading_start() was called but no loading_audio was supplied at construction time, audio was disabled (without_audio=True), or no livekit_config was supplied.
• loading_start() was called but the dedicated "loading-audio" LiveKit track failed to publish or the LiveKit room is not connected.

Any application that already registers an error callback observes loading failures automatically.

LiveKit publisher-timeout reconnect

If the LiveKit room reconnects while the loop was running, the server tears down the loop and clears the audio source. The loop does not auto-resume; if you still want loading audio to play, call loading_start() again after the reconnect.

---

Webhook Receiver

The WebhookReceiver class securely verifies and parses cryptographically signed webhooks from the Sayna SIP service.

Security Features

• HMAC-SHA256 Signature Verification: Ensures webhook authenticity
• Constant-Time Comparison: Prevents timing attack vulnerabilities
• Replay Protection: 5-minute timestamp window prevents replay attacks
• Strict Validation: Comprehensive checks on all required fields

WebhookReceiver(secret=None)

Creates a new webhook receiver instance.

Parameter	Type	Default	Purpose
secret	str (optional)	None	HMAC signing secret (min 16 chars). Falls back to SAYNA_WEBHOOK_SECRET env variable.

Example:

from sayna_client import WebhookReceiver

# Explicit secret
receiver = WebhookReceiver("your-secret-key-min-16-chars")

# Or from environment variable (SAYNA_WEBHOOK_SECRET)
receiver = WebhookReceiver()

---

receiver.receive(headers, body)

Verifies and parses an incoming SIP webhook.

Parameter	Type	Purpose
headers	dict	HTTP request headers (case-insensitive).
body	str	Raw request body as string (not parsed JSON).

Returns: WebhookSIPOutput - Validated webhook payload with fields:

• participant: Object with identity, sid, and optional name
• room: Object with name and sid
• from_phone_number: Caller's phone number (E.164 format)
• to_phone_number: Called phone number (E.164 format)
• room_prefix: Room name prefix configured in Sayna
• sip_host: SIP domain extracted from the To header
• sip_headers: SIP headers from the incoming call (optional, dict[str, str] or None)

Raises: SaynaValidationError if signature verification fails or payload is invalid.

---

Flask Example

from flask import Flask, request, jsonify
from sayna_client import WebhookReceiver, SaynaValidationError

app = Flask(__name__)
receiver = WebhookReceiver("your-secret-key-min-16-chars")

@app.route("/webhook", methods=["POST"])
def webhook():
    try:
        # CRITICAL: Pass raw body, not parsed JSON
        body = request.get_data(as_text=True)
        webhook = receiver.receive(request.headers, body)

        print(f"From: {webhook.from_phone_number}")
        print(f"To: {webhook.to_phone_number}")
        print(f"Room: {webhook.room.name}")
        print(f"SIP Host: {webhook.sip_host}")
        print(f"Participant: {webhook.participant.identity}")

        return jsonify({"received": True}), 200
    except SaynaValidationError as error:
        print(f"Webhook verification failed: {error}")
        return jsonify({"error": "Invalid signature"}), 401

---

FastAPI Example

from fastapi import FastAPI, Request, HTTPException
from sayna_client import WebhookReceiver, SaynaValidationError

app = FastAPI()
receiver = WebhookReceiver()  # Uses SAYNA_WEBHOOK_SECRET env variable

@app.post("/webhook")
async def webhook(request: Request):
    try:
        body = await request.body()
        body_str = body.decode("utf-8")
        webhook = receiver.receive(dict(request.headers), body_str)

        # Process webhook...
        return {"received": True}
    except SaynaValidationError as error:
        raise HTTPException(status_code=401, detail=str(error))

---

Important Notes

• Raw Body Required: You MUST pass the raw request body string, not the parsed JSON object. The signature is computed over the exact bytes received, so any formatting changes will cause verification to fail.

• Case-Insensitive Headers: Header names are case-insensitive in HTTP. The receiver handles both X-Sayna-Signature and x-sayna-signature correctly.

• Secret Security: Never commit secrets to version control. Use environment variables or a secret management system. Generate a secure secret with: openssl rand -hex 32

---

Development

Setup

# Create a virtual environment
python -m venv .venv

# Activate it
source .venv/bin/activate  # On Linux/macOS
# .venv\Scripts\activate  # On Windows

# Install development dependencies
pip install -e ".[dev]"

Tip: For faster dependency installation, you can use uv:

pip install uv
uv pip install -e ".[dev]"

Running Tests

# Run all tests
pytest

# Run with coverage
pytest --cov=sayna_client --cov-report=html

# Run specific test file
pytest tests/test_client.py

Type Checking

mypy src/sayna_client

Linting and Formatting

This project uses Ruff for linting and formatting:

# Check code
ruff check .

# Format code
ruff format .

# Fix auto-fixable issues
ruff check --fix .

Requirements

• Python 3.9 or higher
• aiohttp >= 3.9.0
• pydantic >= 2.0.0

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Support

For issues and questions, please visit the GitHub Issues page.