(phonic) Add and match audio output options: frame size and sample rate

livekit-agents/livekit/agents/voice/room_io/_output.py

@@ -30,6 +30,7 @@ def __init__(
         *,
         sample_rate: int,
         num_channels: int,
+        frame_size_ms: int = 50,
         track_publish_options: rtc.TrackPublishOptions,
         track_name: str = "roomio_audio",
     ) -> None:
@@ -49,7 +50,7 @@ def __init__(
 
         self._audio_buf = utils.aio.Chan[rtc.AudioFrame]()
         self._audio_bstream = utils.audio.AudioByteStream(
-            sample_rate, num_channels, samples_per_channel=sample_rate // 20, progressive=True
+            sample_rate, num_channels, samples_per_channel=sample_rate * frame_size_ms // 1000, progressive=True
         )
 
         self._flush_task: asyncio.Task[None] | None = None

livekit-agents/livekit/agents/voice/room_io/room_io.py

@@ -133,6 +133,7 @@ async def start(self) -> None:
                 self._room,
                 sample_rate=output_audio_options.sample_rate,
                 num_channels=output_audio_options.num_channels,
+                frame_size_ms=output_audio_options.frame_size_ms,
                 track_publish_options=output_audio_options.track_publish_options,
                 track_name=(
                     output_audio_options.track_name

livekit-agents/livekit/agents/voice/room_io/types.py

@@ -83,6 +83,8 @@ class VideoInputOptions:
 class AudioOutputOptions:
     sample_rate: int = 24000
     num_channels: int = 1
+    frame_size_ms: int = 50
+    """The frame size in milliseconds for audio output frames."""
     track_publish_options: rtc.TrackPublishOptions = field(
         default_factory=lambda: rtc.TrackPublishOptions(source=rtc.TrackSource.SOURCE_MICROPHONE)
     )

livekit-plugins/livekit-plugins-phonic/livekit/plugins/phonic/realtime/realtime_model.py

@@ -1,48 +1,50 @@
 from __future__ import annotations
 
 import asyncio
 import base64
 import json
 import os
 import time
 import typing
 import weakref
 from collections.abc import AsyncIterable
 from dataclasses import dataclass, field
 from typing import Literal
 
 from livekit import rtc
 from livekit.agents import llm, utils
+from livekit.agents.voice.room_io import AudioOutputOptions
 from livekit.agents.types import (
     DEFAULT_API_CONNECT_OPTIONS,
     NOT_GIVEN,
     APIConnectOptions,
     NotGivenOr,
 )
 from livekit.agents.utils import audio as audio_utils, is_given
 from phonic import AsyncPhonic
 from phonic.conversations.socket_client import (
     AsyncConversationsSocketClient,
 )
 from phonic.types import (
     AddSystemMessagePayload,
     AudioChunkPayload,
     AudioChunkResponsePayload,
     ConfigPayload,
     GenerateReplyPayload,
     InputTextPayload,
     SayPayload,
     ToolCallInterruptedPayload,
     ToolCallOutputPayload,
     ToolCallPayload,
 )
 
 from ..log import logger
 
 PHONIC_INPUT_SAMPLE_RATE = 44100
 PHONIC_OUTPUT_SAMPLE_RATE = 44100
 PHONIC_NUM_CHANNELS = 1
 PHONIC_INPUT_FRAME_MS = 20
+PHONIC_OUTPUT_FRAME_MS = 20
 WS_CLOSE_NORMAL = 1000
 TOOL_CALL_OUTPUT_TIMEOUT_MS = 60000
 
@@ -221,6 +223,26 @@
         self._sessions.add(sess)
         return sess
 
+    @staticmethod
+    def audio_output_options(**kwargs: typing.Any) -> AudioOutputOptions:
+        """Return ``AudioOutputOptions`` tuned for Phonic: 44100 Hz sample rate and 20 ms frames.
+
+        Using these options with ``RoomIO`` avoids resampling (Phonic natively outputs 44100 Hz)
+        and matches the frame size to Phonic's 20 ms audio chunks, minimising buffer starvation.
+
+        Example::
+
+            model = phonic.realtime.RealtimeModel(...)
+            room_io = RoomIO(session, room=ctx.room, options=RoomOptions(
+                audio_output=model.audio_output_options()
+            ))
+        """
+        return AudioOutputOptions(
+            sample_rate=PHONIC_OUTPUT_SAMPLE_RATE,
+            frame_size_ms=PHONIC_OUTPUT_FRAME_MS,
+            **kwargs,
+        )
+
     def update_options(
         self,
     ) -> None: