livekit-plugins-voho 0.1.0__py3-none-any.whl

livekit/plugins/voho/__init__.py

@@ -0,0 +1,53 @@
+"""Voho for LiveKit Agents: Arabic speech, Saudi and Gulf dialects.
+
+    from livekit.plugins import voho
+
+    session = AgentSession(
+        stt=voho.STT(language="ar-SA"),
+        tts=voho.TTS(voice="layla"),
+        llm=...,
+    )
+
+Set ``VOHO_API_KEY`` or pass ``api_key=``. Keys are created at
+https://app.voho.ai/tokens and billed there. Docs: https://docs.voho.ai/livekit
+"""
+
+from .models import DEFAULT_LANGUAGE, DEFAULT_MODEL, DEFAULT_VOICE, STTLanguage, TTSModel, TTSVoice
+from .stt import STT, RecognizeStream
+from .tts import TTS, ChunkedStream, SynthesizeStream
+from .version import __version__
+
+__all__ = [
+    "STT",
+    "TTS",
+    "RecognizeStream",
+    "ChunkedStream",
+    "SynthesizeStream",
+    "STTLanguage",
+    "TTSModel",
+    "TTSVoice",
+    "DEFAULT_LANGUAGE",
+    "DEFAULT_MODEL",
+    "DEFAULT_VOICE",
+    "__version__",
+]
+
+from livekit.agents import Plugin  # noqa: E402
+
+from .log import logger  # noqa: E402
+
+
+class VohoPlugin(Plugin):
+    def __init__(self) -> None:
+        super().__init__(__name__, __version__, __package__, logger)
+
+
+Plugin.register_plugin(VohoPlugin())
+
+# 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

livekit/plugins/voho/_api.py

@@ -0,0 +1,86 @@
+"""The little that both halves of the plugin share: where the API is, how it
+is authenticated, and how its errors become LiveKit's.
+
+Kept apart from the STT and TTS modules so a change to the error contract is
+one edit. Voho errors are ``{"error": {"code", "message"}}`` with an HTTP
+status, and the code is the part worth surfacing: a developer whose agent
+stopped talking wants to see ``insufficient_credit`` in the log, not a 402.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+import os
+from typing import Any
+
+import aiohttp
+
+from livekit.agents import APIConnectionError, APIStatusError, APITimeoutError
+
+DEFAULT_BASE_URL = "https://app.voho.ai"
+
+
+def resolve_api_key(api_key: str | None) -> str:
+    key = api_key or os.environ.get("VOHO_API_KEY")
+    if not key:
+        raise ValueError(
+            "Voho API key is required: pass api_key=... or set VOHO_API_KEY. "
+            "Create one at https://app.voho.ai/tokens"
+        )
+    return key
+
+
+def resolve_base_url(base_url: str | None) -> str:
+    return (base_url or os.environ.get("VOHO_BASE_URL") or DEFAULT_BASE_URL).rstrip("/")
+
+
+def ws_url(base_url: str, path: str) -> str:
+    if base_url.startswith("https://"):
+        return "wss://" + base_url[len("https://") :] + path
+    if base_url.startswith("http://"):
+        return "ws://" + base_url[len("http://") :] + path
+    return base_url + path
+
+
+def headers(api_key: str) -> dict[str, str]:
+    return {"Authorization": f"Bearer {api_key}", "User-Agent": "livekit-plugins-voho"}
+
+
+def error_from_body(status: int, body: bytes | str) -> APIStatusError:
+    """A Voho error body, turned into the exception LiveKit retries or reports on."""
+    code, message = "http_error", f"HTTP {status}"
+    try:
+        parsed = json.loads(body if isinstance(body, str) else body.decode("utf-8", "replace"))
+        err = parsed.get("error") if isinstance(parsed, dict) else None
+        if isinstance(err, dict):
+            code = str(err.get("code", code))
+            message = str(err.get("message", message))
+    except (ValueError, AttributeError):
+        pass
+    # 4xx other than rate limiting is the caller's to fix; retrying a bad key
+    # or an empty balance just spends the retry budget on the same answer.
+    retryable = status >= 500 or status == 429
+    return APIStatusError(f"voho: {code}: {message}", status_code=status, body=None, retryable=retryable)
+
+
+def error_from_frame(frame: dict[str, Any]) -> APIStatusError:
+    """The ``{"type": "error"}`` frame a Voho socket sends, as an exception."""
+    err = frame.get("error") or {}
+    code = str(err.get("code", "socket_error"))
+    message = str(err.get("message", "The socket reported an error."))
+    terminal = code in {"unauthorized", "insufficient_credit", "unknown_voice", "unknown_model", "unknown_format", "invalid_request", "text_too_long"}
+    return APIStatusError(f"voho: {code}: {message}", status_code=-1, body=None, retryable=not terminal)
+
+
+def translate(exc: BaseException) -> BaseException:
+    """Anything aiohttp or asyncio raises, as the exception LiveKit expects."""
+    if isinstance(exc, (APIStatusError, APIConnectionError, APITimeoutError)):
+        return exc
+    if isinstance(exc, asyncio.TimeoutError):
+        return APITimeoutError()
+    if isinstance(exc, aiohttp.ClientResponseError):
+        return APIStatusError(f"voho: HTTP {exc.status}: {exc.message}", status_code=exc.status, body=None)
+    if isinstance(exc, aiohttp.ClientError):
+        return APIConnectionError(f"voho: {exc}")
+    return APIConnectionError(f"voho: {exc!r}")

livekit/plugins/voho/log.py

@@ -0,0 +1,3 @@
+import logging
+
+logger = logging.getLogger("livekit.plugins.voho")

livekit/plugins/voho/models.py

@@ -0,0 +1,40 @@
+"""Names the API accepts, so a typo is a type error rather than a 400 at runtime.
+
+These mirror the catalogue served by ``GET /v1/voices``. The literal types are
+for editor completion and static checking; any string is still passed through,
+so a voice added to the catalogue after this release works without an upgrade.
+"""
+
+from typing import Literal, Union
+
+TTSModel = Literal["sada-1", "nabra-1"]
+"""``sada-1`` is the flagship and the only tier with a streaming path."""
+
+TTSVoice = Union[
+    Literal[
+        # Najdi — Riyadh and the centre
+        "layla", "nouf", "faisal", "omar",
+        # Hijazi — Jeddah, Makkah, Madinah
+        "salma", "rawan", "hisham", "tariq",
+        # Omani — Muscat and the interior
+        "maryam", "salim",
+        # Gulf, Egyptian, Modern Standard
+        "reem", "maha", "khalid", "yousef",
+        # English
+        "astra", "bancroft", "clementine", "marlow", "vespera", "cupola",
+    ],
+    str,
+]
+
+STTLanguage = Union[Literal["ar-SA", "ar-OM", "ar-AE", "ar-EG", "ar", "en-US", "en-GB"], str]
+"""Any Arabic code also transcribes English mixed into the same sentence."""
+
+DEFAULT_VOICE: TTSVoice = "layla"
+DEFAULT_MODEL: TTSModel = "sada-1"
+DEFAULT_LANGUAGE: STTLanguage = "ar-SA"
+
+TTS_SAMPLE_RATE = 24000
+"""What ``format=pcm`` streams: 16-bit little-endian mono at 24 kHz."""
+
+STT_SAMPLE_RATE = 16000
+"""What the transcription socket is opened at. Input is resampled to this."""

livekit/plugins/voho/stt.py

@@ -0,0 +1,255 @@
+"""Voho speech-to-text for LiveKit Agents.
+
+Arabic with English mixed in, which is how the Gulf speaks on the phone: a
+caller starts in Najdi and says the product name in English, and a
+transcriber locked to one language writes nonsense at exactly that moment.
+Any ``ar-*`` language code here transcribes both.
+
+``recognize(buffer)`` sends a finished utterance to ``POST /v1/transcribe``.
+``stream()`` opens ``/v1/transcribe/ws`` and sends audio frames as they arrive,
+getting interim results back while the person is still talking — which is
+what LiveKit's turn detection and interruption handling are built to consume.
+Input is resampled to 16 kHz mono by the base class before it reaches here.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+from dataclasses import dataclass, replace
+
+import aiohttp
+from livekit import rtc
+
+from livekit.agents import (
+    DEFAULT_API_CONNECT_OPTIONS,
+    APIConnectOptions,
+    stt,
+    utils,
+)
+from livekit.agents.types import NOT_GIVEN, NotGivenOr
+from livekit.agents.utils import AudioBuffer, is_given
+
+from . import _api
+from .log import logger
+from .models import DEFAULT_LANGUAGE, STT_SAMPLE_RATE, STTLanguage
+
+
+@dataclass
+class _Options:
+    language: str
+    base_url: str
+
+
+class STT(stt.STT):
+    def __init__(
+        self,
+        *,
+        language: STTLanguage = DEFAULT_LANGUAGE,
+        api_key: str | None = None,
+        base_url: str | None = None,
+        http_session: aiohttp.ClientSession | None = None,
+    ) -> None:
+        """Voho STT.
+
+        Args:
+            language: ``ar-SA`` by default. Any Arabic code also picks up English
+                in the same sentence. ``en-US`` for an English-only line.
+            api_key: A ``voho_sk_live_…`` key, or ``VOHO_API_KEY``.
+            base_url: Override for a private or Saudi-hosted deployment, or
+                ``VOHO_BASE_URL``.
+            http_session: Share LiveKit's session rather than opening one.
+        """
+        super().__init__(capabilities=stt.STTCapabilities(streaming=True, interim_results=True))
+        self._api_key = _api.resolve_api_key(api_key)
+        self._opts = _Options(language=language, base_url=_api.resolve_base_url(base_url))
+        self._session = http_session
+
+    @property
+    def model(self) -> str:
+        return "voho-transcribe"
+
+    @property
+    def provider(self) -> str:
+        return "Voho"
+
+    def _ensure_session(self) -> aiohttp.ClientSession:
+        if not self._session:
+            self._session = utils.http_context.http_session()
+        return self._session
+
+    def update_options(self, *, language: NotGivenOr[STTLanguage] = NOT_GIVEN) -> None:
+        if is_given(language):
+            self._opts = replace(self._opts, language=language)
+
+    async def _recognize_impl(
+        self,
+        buffer: AudioBuffer,
+        *,
+        language: NotGivenOr[str] = NOT_GIVEN,
+        conn_options: APIConnectOptions,
+    ) -> stt.SpeechEvent:
+        lang = language if is_given(language) else self._opts.language
+        wav = rtc.combine_audio_frames(buffer).to_wav_bytes()
+
+        form = aiohttp.FormData()
+        form.add_field("file", wav, filename="utterance.wav", content_type="audio/wav")
+        form.add_field("language", lang)
+
+        try:
+            async with self._ensure_session().post(
+                f"{self._opts.base_url}/v1/transcribe",
+                headers=_api.headers(self._api_key),
+                data=form,
+                timeout=aiohttp.ClientTimeout(total=60, sock_connect=conn_options.timeout),
+            ) as resp:
+                if resp.status >= 400:
+                    raise _api.error_from_body(resp.status, await resp.read())
+                body = await resp.json()
+        except BaseException as exc:
+            raise _api.translate(exc) from exc
+
+        return stt.SpeechEvent(
+            type=stt.SpeechEventType.FINAL_TRANSCRIPT,
+            request_id=utils.shortuuid("voho_"),
+            alternatives=[
+                stt.SpeechData(
+                    language=lang,
+                    text=str(body.get("text") or ""),
+                    confidence=float(body.get("confidence") or 0.0),
+                )
+            ],
+        )
+
+    def stream(
+        self,
+        *,
+        language: NotGivenOr[str] = NOT_GIVEN,
+        conn_options: APIConnectOptions = DEFAULT_API_CONNECT_OPTIONS,
+    ) -> RecognizeStream:
+        return RecognizeStream(
+            stt=self,
+            language=language if is_given(language) else self._opts.language,
+            conn_options=conn_options,
+        )
+
+    async def aclose(self) -> None:
+        pass
+
+
+class RecognizeStream(stt.RecognizeStream):
+    """Frames in, transcripts out, on one socket for the life of the stream."""
+
+    def __init__(self, *, stt: STT, language: str, conn_options: APIConnectOptions) -> None:
+        super().__init__(stt=stt, conn_options=conn_options, sample_rate=STT_SAMPLE_RATE)
+        self._stt: STT = stt
+        self._language = language
+
+    async def _run(self) -> None:
+        opts = self._stt._opts
+        request_id = utils.shortuuid("voho_")
+        ws: aiohttp.ClientWebSocketResponse | None = None
+
+        try:
+            ws = await asyncio.wait_for(
+                self._stt._ensure_session().ws_connect(
+                    _api.ws_url(opts.base_url, "/v1/transcribe/ws"),
+                    headers=_api.headers(self._stt._api_key),
+                ),
+                self._conn_options.timeout,
+            )
+            await _expect(ws, "ready")
+            await ws.send_str(
+                json.dumps(
+                    {"type": "start", "language": self._language, "sample_rate": STT_SAMPLE_RATE, "encoding": "pcm"}
+                )
+            )
+            await _expect(ws, "started")
+
+            async def send() -> None:
+                assert ws is not None
+                async for item in self._input_ch:
+                    if isinstance(item, self._FlushSentinel):
+                        # A flush is a hint that an utterance has ended. The
+                        # server segments on silence itself, so there is
+                        # nothing to send; the final result follows on its own.
+                        continue
+                    await ws.send_bytes(item.data.tobytes())
+                await ws.send_str(json.dumps({"type": "stop"}))
+
+            async def receive() -> None:
+                assert ws is not None
+                speaking = False
+                async for msg in ws:
+                    if msg.type != aiohttp.WSMsgType.TEXT:
+                        if msg.type in (aiohttp.WSMsgType.CLOSED, aiohttp.WSMsgType.CLOSE, aiohttp.WSMsgType.ERROR):
+                            break
+                        continue
+                    frame = json.loads(msg.data)
+                    kind = frame.get("type")
+                    if kind == "transcript":
+                        text = str(frame.get("text") or "")
+                        if not text:
+                            continue
+                        final = bool(frame.get("final"))
+                        data = stt.SpeechData(
+                            language=str(frame.get("language") or self._language),
+                            text=text,
+                            confidence=float(frame.get("confidence") or 0.0),
+                        )
+                        if not speaking:
+                            speaking = True
+                            self._event_ch.send_nowait(
+                                stt.SpeechEvent(type=stt.SpeechEventType.START_OF_SPEECH, request_id=request_id)
+                            )
+                        self._event_ch.send_nowait(
+                            stt.SpeechEvent(
+                                type=stt.SpeechEventType.FINAL_TRANSCRIPT if final else stt.SpeechEventType.INTERIM_TRANSCRIPT,
+                                request_id=request_id,
+                                alternatives=[data],
+                            )
+                        )
+                        if final:
+                            speaking = False
+                            self._event_ch.send_nowait(
+                                stt.SpeechEvent(type=stt.SpeechEventType.END_OF_SPEECH, request_id=request_id)
+                            )
+                    elif kind == "done":
+                        return
+                    elif kind == "error":
+                        raise _api.error_from_frame(frame)
+                    else:
+                        logger.debug("voho stt: ignoring frame %s", kind)
+                # The server closed without saying done; if we were still
+                # sending, that is a dropped connection and LiveKit may retry.
+                if not self._input_ch.closed:
+                    raise aiohttp.ClientConnectionError("voho: transcription socket closed unexpectedly")
+
+            send_task = asyncio.create_task(send(), name="voho-stt-send")
+            recv_task = asyncio.create_task(receive(), name="voho-stt-recv")
+            try:
+                await asyncio.gather(send_task, recv_task)
+            finally:
+                for t in (send_task, recv_task):
+                    if not t.done():
+                        t.cancel()
+        except BaseException as exc:
+            raise _api.translate(exc) from exc
+        finally:
+            if ws is not None:
+                try:
+                    await ws.close()
+                except Exception:
+                    pass
+
+
+async def _expect(ws: aiohttp.ClientWebSocketResponse, kind: str) -> dict:
+    async for msg in ws:
+        if msg.type != aiohttp.WSMsgType.TEXT:
+            continue
+        frame = json.loads(msg.data)
+        if frame.get("type") == kind:
+            return frame
+        if frame.get("type") == "error":
+            raise _api.error_from_frame(frame)
+    raise aiohttp.ClientConnectionError(f"voho: socket closed while waiting for {kind}")

livekit/plugins/voho/tts.py

@@ -0,0 +1,300 @@
+"""Voho text-to-speech for LiveKit Agents.
+
+Two paths, because LiveKit asks for two:
+
+``synthesize(text)`` has the whole sentence up front and wants audio back as
+fast as it can be produced. That is ``POST /v1/speech/stream`` with a raw
+PCM body: chunked HTTP, first audio in roughly 200 ms.
+
+``stream()`` is fed tokens as an LLM emits them and must start speaking before
+the sentence is finished. That is ``/v1/speech/ws``: text frames in, audio
+frames out, pipelined rather than batched at the end.
+
+Sockets are pooled. Opening one costs a TLS handshake and a key check, several
+hundred milliseconds that a caller hears as the agent hesitating before every
+sentence; once open, first audio follows the text in about 200 ms. So a socket
+that has finished an utterance goes back to the pool for the next, ``prewarm()``
+opens one before the first sentence is needed, and a socket idle for longer
+than the proxy in front of the API allows is replaced rather than reused.
+
+Audio is 16-bit little-endian mono PCM at 24 kHz in both cases. LiveKit
+resamples for the room, so nothing here does.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+from dataclasses import dataclass, replace
+
+import aiohttp
+
+from livekit.agents import (
+    DEFAULT_API_CONNECT_OPTIONS,
+    APIConnectOptions,
+    tts,
+    utils,
+)
+from livekit.agents.types import NOT_GIVEN, NotGivenOr
+from livekit.agents.utils import is_given
+
+from . import _api
+from .log import logger
+from .models import DEFAULT_MODEL, DEFAULT_VOICE, TTS_SAMPLE_RATE, TTSModel, TTSVoice
+
+
+@dataclass
+class _Options:
+    voice: str
+    model: str
+    base_url: str
+
+
+class TTS(tts.TTS):
+    def __init__(
+        self,
+        *,
+        voice: TTSVoice = DEFAULT_VOICE,
+        model: TTSModel = DEFAULT_MODEL,
+        api_key: str | None = None,
+        base_url: str | None = None,
+        http_session: aiohttp.ClientSession | None = None,
+    ) -> None:
+        """Voho TTS.
+
+        Args:
+            voice: A voice id from ``GET /v1/voices`` — ``layla`` (Najdi) by
+                default; ``salma`` for Hijazi, ``maryam`` for Omani, ``clementine``
+                for English. The dialect is the voice.
+            model: ``sada-1`` (default, streaming) or ``nabra-1`` (cheaper,
+                chunked only). Passing ``nabra-1`` makes ``stream()`` fall back to
+                per-sentence synthesis.
+            api_key: A ``voho_sk_live_…`` key, or ``VOHO_API_KEY``.
+            base_url: Override for a private or Saudi-hosted deployment, or
+                ``VOHO_BASE_URL``.
+            http_session: Share LiveKit's session rather than opening one.
+        """
+        super().__init__(
+            capabilities=tts.TTSCapabilities(streaming=model == "sada-1"),
+            sample_rate=TTS_SAMPLE_RATE,
+            num_channels=1,
+        )
+        self._api_key = _api.resolve_api_key(api_key)
+        self._opts = _Options(voice=voice, model=model, base_url=_api.resolve_base_url(base_url))
+        self._session = http_session
+        self._pool = utils.ConnectionPool[aiohttp.ClientWebSocketResponse](
+            connect_cb=self._connect_ws,
+            close_cb=self._close_ws,
+            # Idle sockets are cut by the proxy at 120 s; refresh on use, so a
+            # socket in steady use lives on and one left idle is replaced.
+            max_session_duration=50,
+            mark_refreshed_on_get=True,
+        )
+
+    async def _connect_ws(self, timeout: float) -> aiohttp.ClientWebSocketResponse:
+        ws = await asyncio.wait_for(
+            self._ensure_session().ws_connect(
+                _api.ws_url(self._opts.base_url, "/v1/speech/ws"),
+                headers=_api.headers(self._api_key),
+            ),
+            timeout,
+        )
+        try:
+            await asyncio.wait_for(_expect(ws, "ready"), timeout)
+        except BaseException:
+            await ws.close()
+            raise
+        return ws
+
+    async def _close_ws(self, ws: aiohttp.ClientWebSocketResponse) -> None:
+        await ws.close()
+
+    def prewarm(self) -> None:
+        """Open a socket before the first sentence, so it is not paid for then."""
+        if self._opts.model == "sada-1":
+            self._pool.prewarm()
+
+    @property
+    def model(self) -> str:
+        return self._opts.model
+
+    @property
+    def provider(self) -> str:
+        return "Voho"
+
+    def _ensure_session(self) -> aiohttp.ClientSession:
+        if not self._session:
+            self._session = utils.http_context.http_session()
+        return self._session
+
+    def update_options(
+        self,
+        *,
+        voice: NotGivenOr[TTSVoice] = NOT_GIVEN,
+        model: NotGivenOr[TTSModel] = NOT_GIVEN,
+    ) -> None:
+        """Change the voice or model for subsequent requests. Switching dialect
+        mid-call — a Hijazi caller on a Najdi line — is this and nothing else."""
+        if is_given(voice):
+            self._opts = replace(self._opts, voice=voice)
+        if is_given(model):
+            self._opts = replace(self._opts, model=model)
+
+    def synthesize(
+        self, text: str, *, conn_options: APIConnectOptions = DEFAULT_API_CONNECT_OPTIONS
+    ) -> ChunkedStream:
+        return ChunkedStream(tts=self, input_text=text, conn_options=conn_options)
+
+    def stream(
+        self, *, conn_options: APIConnectOptions = DEFAULT_API_CONNECT_OPTIONS
+    ) -> SynthesizeStream:
+        if self._opts.model != "sada-1":
+            # The economy tier has no streaming path upstream; LiveKit's own
+            # adapter turns sentence-at-a-time synthesis into a stream.
+            return super().stream(conn_options=conn_options)  # type: ignore[return-value]
+        return SynthesizeStream(tts=self, conn_options=conn_options)
+
+    async def aclose(self) -> None:
+        # The pool is ours; the HTTP session is LiveKit's or the caller's.
+        await self._pool.aclose()
+
+
+class ChunkedStream(tts.ChunkedStream):
+    """Whole text in, chunked PCM out."""
+
+    def __init__(self, *, tts: TTS, input_text: str, conn_options: APIConnectOptions) -> None:
+        super().__init__(tts=tts, input_text=input_text, conn_options=conn_options)
+        self._tts: TTS = tts
+
+    async def _run(self, output_emitter: tts.AudioEmitter) -> None:
+        opts = self._tts._opts
+        request_id = utils.shortuuid("voho_")
+        try:
+            async with self._tts._ensure_session().post(
+                f"{opts.base_url}/v1/speech/stream",
+                headers={**_api.headers(self._tts._api_key), "Accept": "audio/L16"},
+                json={"text": self._input_text, "voice": opts.voice, "model": opts.model, "format": "pcm"},
+                timeout=aiohttp.ClientTimeout(total=60, sock_connect=self._conn_options.timeout),
+            ) as resp:
+                if resp.status >= 400:
+                    raise _api.error_from_body(resp.status, await resp.read())
+
+                output_emitter.initialize(
+                    request_id=request_id,
+                    sample_rate=self._tts.sample_rate,
+                    num_channels=1,
+                    mime_type="audio/pcm",
+                    frame_size_ms=50,
+                )
+                async for data, _ in resp.content.iter_chunks():
+                    output_emitter.push(data)
+                output_emitter.flush()
+        except BaseException as exc:
+            raise _api.translate(exc) from exc
+
+
+class SynthesizeStream(tts.SynthesizeStream):
+    """Tokens in as they arrive, PCM out as it is produced, on a pooled socket."""
+
+    def __init__(self, *, tts: TTS, conn_options: APIConnectOptions) -> None:
+        super().__init__(tts=tts, conn_options=conn_options)
+        self._tts: TTS = tts
+
+    async def _run(self, output_emitter: tts.AudioEmitter) -> None:
+        output_emitter.initialize(
+            request_id=utils.shortuuid("voho_"),
+            sample_rate=self._tts.sample_rate,
+            num_channels=1,
+            mime_type="audio/pcm",
+            # LiveKit holds 200 ms of audio before its first frame by default.
+            # On a phone line that is 200 ms of silence before every sentence.
+            frame_size_ms=50,
+            stream=True,
+        )
+        pool = self._tts._pool
+
+        async def acquire() -> aiohttp.ClientWebSocketResponse:
+            # A pooled socket can have been closed by the far end while it sat
+            # idle; that one is discarded and the next is tried.
+            for _ in range(3):
+                ws = await pool.get(timeout=self._conn_options.timeout)
+                if not ws.closed:
+                    return ws
+                pool.remove(ws)
+            raise aiohttp.ClientConnectionError("voho: could not obtain an open speech socket")
+
+        async def utterance(first: str) -> bool:
+            """Speak one segment. Returns False when input ended without a flush."""
+            opts = self._tts._opts
+            ws = await acquire()
+            ok = False
+            try:
+                await ws.send_str(json.dumps({"type": "start", "voice": opts.voice, "model": opts.model, "format": "pcm"}))
+                await _expect(ws, "started")
+                output_emitter.start_segment(segment_id=utils.shortuuid("seg_"))
+                self._mark_started()
+
+                async def receive() -> None:
+                    async for msg in ws:
+                        if msg.type == aiohttp.WSMsgType.BINARY:
+                            output_emitter.push(msg.data)
+                        elif msg.type == aiohttp.WSMsgType.TEXT:
+                            frame = json.loads(msg.data)
+                            if frame.get("type") == "done":
+                                return
+                            if frame.get("type") == "error":
+                                raise _api.error_from_frame(frame)
+                        elif msg.type in (aiohttp.WSMsgType.CLOSED, aiohttp.WSMsgType.CLOSE, aiohttp.WSMsgType.ERROR):
+                            break
+                    raise aiohttp.ClientConnectionError("voho: socket closed before the utterance finished")
+
+                recv = asyncio.create_task(receive(), name="voho-tts-recv")
+                try:
+                    await ws.send_str(json.dumps({"type": "text", "text": first}))
+                    flushed = False
+                    async for item in self._input_ch:
+                        if isinstance(item, self._FlushSentinel):
+                            flushed = True
+                            break
+                        if item:
+                            await ws.send_str(json.dumps({"type": "text", "text": item}))
+                        if recv.done():
+                            recv.result()  # surface a receive-side failure now
+                    await ws.send_str(json.dumps({"type": "flush"}))
+                    await recv
+                finally:
+                    if not recv.done():
+                        recv.cancel()
+                output_emitter.end_segment()
+                ok = True
+                return flushed
+            finally:
+                # A socket that finished cleanly goes back for the next
+                # sentence; anything else is closed, never reused half-way.
+                if ok:
+                    pool.put(ws)
+                else:
+                    pool.remove(ws)
+
+        try:
+            async for item in self._input_ch:
+                if isinstance(item, self._FlushSentinel) or not item:
+                    continue
+                if not await utterance(item):
+                    break
+        except BaseException as exc:
+            raise _api.translate(exc) from exc
+
+
+async def _expect(ws: aiohttp.ClientWebSocketResponse, kind: str) -> dict:
+    """Read control frames until the one we are waiting for, raising on error."""
+    async for msg in ws:
+        if msg.type != aiohttp.WSMsgType.TEXT:
+            continue
+        frame = json.loads(msg.data)
+        if frame.get("type") == kind:
+            return frame
+        if frame.get("type") == "error":
+            raise _api.error_from_frame(frame)
+        logger.debug("voho tts: ignoring frame %s while waiting for %s", frame.get("type"), kind)
+    raise aiohttp.ClientConnectionError(f"voho: socket closed while waiting for {kind}")

livekit/plugins/voho/version.py

@@ -0,0 +1 @@
+__version__ = "0.1.0"

livekit_plugins_voho-0.1.0.dist-info/METADATA

@@ -0,0 +1,152 @@
+Metadata-Version: 2.5
+Name: livekit-plugins-voho
+Version: 0.1.0
+Summary: Saudi Arabic voice for LiveKit Agents: Najdi and Hijazi accents, Arabic-English code-switching, streaming speech-to-text and text-to-speech.
+Project-URL: Documentation, https://docs.voho.ai/livekit
+Project-URL: Website, https://voho.ai
+Project-URL: Source, https://github.com/yar-malik/livekit-plugins-voho
+Project-URL: Issues, https://github.com/yar-malik/livekit-plugins-voho/issues
+Author-email: Voho <hello@voho.ai>
+License-Expression: Apache-2.0
+License-File: LICENSE
+Keywords: arabic,arabic stt,arabic tts,gulf arabic,hijazi,ksa,livekit,livekit-agents,najdi,saudi accent,saudi arabic,speech-to-text,text-to-speech,voice-agent
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Natural Language :: Arabic
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Topic :: Multimedia :: Sound/Audio
+Classifier: Topic :: Multimedia :: Video
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.9
+Requires-Dist: aiohttp>=3.9
+Requires-Dist: livekit-agents>=1.0
+Provides-Extra: dev
+Requires-Dist: numpy; extra == 'dev'
+Requires-Dist: pytest-asyncio>=0.23; extra == 'dev'
+Requires-Dist: pytest>=8; extra == 'dev'
+Description-Content-Type: text/markdown
+
+# livekit-plugins-voho
+
+**Saudi Arabic voice for LiveKit Agents.** Speech-to-text and text-to-speech that sound like the Kingdom answers the phone: Najdi for Riyadh, Hijazi for Jeddah and Makkah, plus Omani, Gulf, Egyptian and Modern Standard Arabic. It also handles English mixed into the same sentence, because that is how Saudi callers actually talk.
+
+```bash
+pip install livekit-plugins-voho
+```
+
+```python
+from livekit.agents import AgentSession
+from livekit.plugins import voho
+
+session = AgentSession(
+    stt=voho.STT(language="ar-SA"),   # Saudi Arabic, with English mixed in
+    tts=voho.TTS(voice="layla"),      # a Najdi voice from Riyadh
+    llm=...,                          # any LiveKit LLM plugin
+    vad=...,                          # e.g. silero
+)
+```
+
+Get a key at [app.voho.ai](https://app.voho.ai), then `export VOHO_API_KEY=voho_sk_live_...`.
+
+---
+
+## Why a Saudi-specific plugin
+
+Most "Arabic" voices are Modern Standard Arabic: the register of the news, not of a phone call. A Riyadh caller hears it as a recording. A Jeddah caller hears a Najdi agent as out of town, the way a London voice sounds in Glasgow. The dialect is not a detail. It is whether the caller stays on the line.
+
+Voho is built around that:
+
+- **Saudi dialects as separate voices.** Pick Najdi or Hijazi by name, and switch mid-call when a Jeddah caller rings a Riyadh line.
+- **Arabic and English in one sentence.** "أبغى أغيّر الـ delivery address" is transcribed as said, rather than turned into nonsense at the switch.
+- **Streaming both ways.** Text is spoken while the LLM is still writing the sentence, and transcripts arrive while the caller is still talking, so LiveKit's turn-taking and barge-in work as designed.
+- **Telephone-ready.** Put a LiveKit SIP trunk in front of it and it answers a Saudi 9200, 800 or geographic number.
+- **Saudi hosting for enterprise.** The same plugin points at an in-Kingdom or on-premise Voho deployment with one `base_url` change.
+
+## Voices
+
+The dialect is the voice.
+
+| Voice | Dialect | Where it sounds local | Good for |
+| --- | --- | --- | --- |
+| `layla` | Najdi, female | Riyadh, Qassim, the centre | Reception, appointments (default) |
+| `nouf` | Najdi, female | Riyadh | Collections, compliance, escalations |
+| `faisal` | Najdi, male | Riyadh | Banking, government, long policy text |
+| `omar` | Najdi, male | Riyadh | Outbound confirmations, offers |
+| `salma` | Hijazi, female | Jeddah, Makkah, Madinah | Everyday customer service |
+| `rawan` | Hijazi, female | Jeddah | Retail, delivery |
+| `hisham` | Hijazi, male | Jeddah | Banking, insurance |
+| `tariq` | Hijazi, male | Jeddah | Bookings, follow-ups |
+| `maryam` | Omani, female | Muscat, the interior | Energy, logistics, marine |
+| `salim` | Omani, male | Muscat | Fleet, retail |
+| `reem` | Gulf, female | UAE | Light, conversational |
+| `maha` | Egyptian, female | Egypt | Reassuring, unhurried |
+| `khalid`, `yousef` | Modern Standard | Region-neutral | Announcements, IVR |
+| `clementine`, `astra`, `marlow`, `vespera` | English | | Expat and international lines |
+
+```python
+tts.update_options(voice="salma")      # a Jeddah caller on a Riyadh line
+stt.update_options(language="en-US")   # an English-only stretch
+```
+
+**The voice sets the sound. The words set the dialect.** Tell your LLM which dialect to write in, or a Najdi voice will be reading Fusha. A Najdi agent says الحين, وش, زين and أبشر; a Hijazi one says دحين, إيش, إيوه and never says أبشر. [`examples/agent.py`](examples/agent.py) has a Najdi instruction that works.
+
+## A Saudi customer-service agent in one file
+
+[`examples/agent.py`](examples/agent.py) is a complete agent for a Riyadh retailer: it speaks Najdi, follows the caller into English and back, stops when interrupted, and looks up an order with a tool mid-call.
+
+```bash
+pip install "livekit-agents[openai,silero]" livekit-plugins-voho
+export VOHO_API_KEY=... OPENAI_API_KEY=...
+python examples/agent.py console
+```
+
+To answer a phone number, create a LiveKit SIP inbound trunk and dispatch rule. The [LiveKit guide](https://docs.voho.ai/livekit) walks through it.
+
+## What the plugin calls
+
+| Method | Voho API | |
+| --- | --- | --- |
+| `TTS.synthesize()` | `POST /v1/speech/stream` | Whole sentence in, chunked PCM out |
+| `TTS.stream()` | `WS /v1/speech/ws` | Tokens in as the LLM writes, audio out as it is produced |
+| `STT.recognize()` | `POST /v1/transcribe` | One utterance |
+| `STT.stream()` | `WS /v1/transcribe/ws` | Interim and final transcripts while the caller speaks |
+
+Audio is 16-bit mono PCM: 24 kHz from the TTS, 16 kHz into the STT. LiveKit resamples for the room.
+
+## Pricing
+
+| | |
+| --- | --- |
+| Text-to-speech (`sada-1`, streaming) | 5¢ per 1,000 characters |
+| Text-to-speech (`nabra-1`) | 2¢ per 1,000 characters |
+| Speech-to-text | 3¢ per started minute |
+
+Prepaid, per key. `GET /v1/usage` returns what a key used this month, in the units it was billed in.
+
+## Data and hosting
+
+- Audio is not stored. Billing keeps character and minute counts, nothing else.
+- The public API runs in London.
+- In-Kingdom and on-premise deployments are available for enterprise contracts, with the same API and plugin. See [docs.voho.ai/on-prem/residency](https://docs.voho.ai/on-prem/residency).
+
+## Errors
+
+Voho error codes appear in the exception message (`unauthorized`, `insufficient_credit`, `unknown_voice`, `text_too_long`), so the log says what to fix. Client errors are not retried. Server errors and rate limits are, using LiveKit's `APIConnectOptions`.
+
+## Development
+
+```bash
+pip install -e ".[dev]"
+pytest                               # against a local mock of the API, no key needed
+VOHO_API_KEY=... pytest -m live      # real synthesis, and a speak-then-transcribe round trip in Arabic
+```
+
+## Links
+
+- LiveKit guide: [docs.voho.ai/livekit](https://docs.voho.ai/livekit)
+- API reference: [docs.voho.ai](https://docs.voho.ai)
+- Hear the voices: [voho.ai/demos](https://voho.ai/demos)
+- Contact: support@voho.ai
+
+Apache-2.0

livekit_plugins_voho-0.1.0.dist-info/RECORD

@@ -0,0 +1,12 @@
+livekit/plugins/voho/__init__.py,sha256=vqg1zMQbm_1ymFENQ9b4psCeJLTYGhd3Qpoc86hTFiY,1274
+livekit/plugins/voho/_api.py,sha256=qi7xTX-BvhfHssjBCZiOafIGn-kCe2qswgsGVpZuqoY,3579
+livekit/plugins/voho/log.py,sha256=D-_pO223s83CTrzuRvATMF0YPOiY9qCeHNBsofOoJLs,67
+livekit/plugins/voho/models.py,sha256=gFdvML5EoxD437vX9rCijQ26j-9d5tXajdgu0G6vpys,1430
+livekit/plugins/voho/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+livekit/plugins/voho/stt.py,sha256=TBFMHMKwwJZlIsyZs-awvE4D0uWPRXemAbJQm5hGqFw,9938
+livekit/plugins/voho/tts.py,sha256=HtCTHboXamQL6_nqa0pHHKcZlcopp3z9Htvcc3TJAg0,12229
+livekit/plugins/voho/version.py,sha256=kUR5RAFc7HCeiqdlX36dZOHkUI5wI6V_43RpEcD8b-0,22
+livekit_plugins_voho-0.1.0.dist-info/METADATA,sha256=HZIC3S0fncFvDXp2UWDnn3ShtXQAa4mJgoZXJhEK87Q,7432
+livekit_plugins_voho-0.1.0.dist-info/WHEEL,sha256=zOwg4jB6zX2kU910N-cMawjivD6tO8NEWvE12je1bVk,87
+livekit_plugins_voho-0.1.0.dist-info/licenses/LICENSE,sha256=z8d0m5b2O9McPEK1xHG_dWgUBT6EfBDz6wA0F7xSPTA,11358
+livekit_plugins_voho-0.1.0.dist-info/RECORD,,

livekit_plugins_voho-0.1.0.dist-info/WHEEL

@@ -0,0 +1,4 @@
+Wheel-Version: 1.0
+Generator: hatchling 1.32.0
+Root-Is-Purelib: true
+Tag: py3-none-any

livekit_plugins_voho-0.1.0.dist-info/licenses/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   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.