""" The MIT License (MIT) Copyright (c) 2015-2021 Rapptz Copyright (c) 2021-present Pycord Development Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. Some documentation to refer to: - Our main web socket (mWS) sends opcode 4 with a guild ID and channel ID. - The mWS receives VOICE_STATE_UPDATE and VOICE_SERVER_UPDATE. - We pull the session_id from VOICE_STATE_UPDATE. - We pull the token, endpoint and server_id from VOICE_SERVER_UPDATE. - Then we initiate the voice web socket (vWS) pointing to the endpoint. - We send opcode 0 with the user_id, server_id, session_id and token using the vWS. - The vWS sends back opcode 2 with an ssrc, port, modes(array) and heartbeat_interval. - We send a UDP discovery packet to endpoint:port and receive our IP and our port in LE. - Then we send our IP and port via vWS with opcode 1. - When that's all done, we receive opcode 4 from the vWS. - Finally we can transmit data to endpoint:port. """ from __future__ import annotations import asyncio import logging import select import socket import struct import threading import time from typing import TYPE_CHECKING, Any, Callable, Literal, overload from . import opus, utils from .backoff import ExponentialBackoff from .errors import ClientException, ConnectionClosed from .gateway import * from .player import AudioPlayer, AudioSource from .sinks import RawData, RecordingException, Sink from .utils import MISSING if TYPE_CHECKING: from . import abc from .client import Client from .guild import Guild from .opus import Encoder from .state import ConnectionState from .types.voice import GuildVoiceState as GuildVoiceStatePayload from .types.voice import SupportedModes from .types.voice import VoiceServerUpdate as VoiceServerUpdatePayload from .user import ClientUser has_nacl: bool try: import nacl.secret # type: ignore has_nacl = True except ImportError: has_nacl = False __all__ = ( "VoiceProtocol", "VoiceClient", ) _log = logging.getLogger(__name__) class VoiceProtocol: """A class that represents the Discord voice protocol. This is an abstract class. The library provides a concrete implementation under :class:`VoiceClient`. This class allows you to implement a protocol to allow for an external method of sending voice, such as Lavalink_ or a native library implementation. These classes are passed to :meth:`abc.Connectable.connect `. .. _Lavalink: https://github.com/freyacodes/Lavalink Parameters ---------- client: :class:`Client` The client (or its subclasses) that started the connection request. channel: :class:`abc.Connectable` The voice channel that is being connected to. """ def __init__(self, client: Client, channel: abc.Connectable) -> None: self.client: Client = client self.channel: abc.Connectable = channel async def on_voice_state_update(self, data: GuildVoiceStatePayload) -> None: """|coro| An abstract method that is called when the client's voice state has changed. This corresponds to ``VOICE_STATE_UPDATE``. Parameters ---------- data: :class:`dict` The raw `voice state payload`__. .. _voice_state_update_payload: https://discord.com/developers/docs/resources/voice#voice-state-object __ voice_state_update_payload_ """ raise NotImplementedError async def on_voice_server_update(self, data: VoiceServerUpdatePayload) -> None: """|coro| An abstract method that is called when initially connecting to voice. This corresponds to ``VOICE_SERVER_UPDATE``. Parameters ---------- data: :class:`dict` The raw `voice server update payload`__. .. _voice_server_update_payload: https://discord.com/developers/docs/topics/gateway#voice-server-update-voice-server-update-event-fields __ voice_server_update_payload_ """ raise NotImplementedError async def connect(self, *, timeout: float, reconnect: bool) -> None: """|coro| An abstract method called when the client initiates the connection request. When a connection is requested initially, the library calls the constructor under ``__init__`` and then calls :meth:`connect`. If :meth:`connect` fails at some point then :meth:`disconnect` is called. Within this method, to start the voice connection flow it is recommended to use :meth:`Guild.change_voice_state` to start the flow. After which, :meth:`on_voice_server_update` and :meth:`on_voice_state_update` will be called. The order that these two are called is unspecified. Parameters ---------- timeout: :class:`float` The timeout for the connection. reconnect: :class:`bool` Whether reconnection is expected. """ raise NotImplementedError async def disconnect(self, *, force: bool) -> None: """|coro| An abstract method called when the client terminates the connection. See :meth:`cleanup`. Parameters ---------- force: :class:`bool` Whether the disconnection was forced. """ raise NotImplementedError def cleanup(self) -> None: """This method *must* be called to ensure proper clean-up during a disconnect. It is advisable to call this from within :meth:`disconnect` when you are completely done with the voice protocol instance. This method removes it from the internal state cache that keeps track of currently alive voice clients. Failure to clean-up will cause subsequent connections to report that it's still connected. """ key_id, _ = self.channel._get_voice_client_key() self.client._connection._remove_voice_client(key_id) class VoiceClient(VoiceProtocol): """Represents a Discord voice connection. You do not create these, you typically get them from e.g. :meth:`VoiceChannel.connect`. Attributes ---------- session_id: :class:`str` The voice connection session ID. token: :class:`str` The voice connection token. endpoint: :class:`str` The endpoint we are connecting to. channel: :class:`abc.Connectable` The voice channel connected to. loop: :class:`asyncio.AbstractEventLoop` The event loop that the voice client is running on. Warning ------- In order to use PCM based AudioSources, you must have the opus library installed on your system and loaded through :func:`opus.load_opus`. Otherwise, your AudioSources must be opus encoded (e.g. using :class:`FFmpegOpusAudio`) or the library will not be able to transmit audio. """ endpoint_ip: str voice_port: int secret_key: list[int] ssrc: int def __init__(self, client: Client, channel: abc.Connectable): if not has_nacl: raise RuntimeError("PyNaCl library needed in order to use voice") super().__init__(client, channel) state = client._connection self.token: str = MISSING self.socket = MISSING self.loop: asyncio.AbstractEventLoop = state.loop self._state: ConnectionState = state # this will be used in the AudioPlayer thread self._connected: threading.Event = threading.Event() self._handshaking: bool = False self._potentially_reconnecting: bool = False self._voice_state_complete: asyncio.Event = asyncio.Event() self._voice_server_complete: asyncio.Event = asyncio.Event() self.mode: str = MISSING self._connections: int = 0 self.sequence: int = 0 self.timestamp: int = 0 self.timeout: float = 0 self._runner: asyncio.Task = MISSING self._player: AudioPlayer | None = None self.encoder: Encoder = MISSING self.decoder = None self._lite_nonce: int = 0 self.ws: DiscordVoiceWebSocket = MISSING self.paused = False self.recording = False self.user_timestamps = {} self.sink = None self.starting_time = None self.stopping_time = None warn_nacl = not has_nacl supported_modes: tuple[SupportedModes, ...] = ( "xsalsa20_poly1305_lite", "xsalsa20_poly1305_suffix", "xsalsa20_poly1305", ) @property def guild(self) -> Guild | None: """The guild we're connected to, if applicable.""" return getattr(self.channel, "guild", None) @property def user(self) -> ClientUser: """The user connected to voice (i.e. ourselves).""" return self._state.user def checked_add(self, attr, value, limit): val = getattr(self, attr) if val + value > limit: setattr(self, attr, 0) else: setattr(self, attr, val + value) # connection related async def on_voice_state_update(self, data: GuildVoiceStatePayload) -> None: self.session_id = data["session_id"] channel_id = data["channel_id"] if not self._handshaking or self._potentially_reconnecting: # If we're done handshaking then we just need to update ourselves # If we're potentially reconnecting due to a 4014, then we need to differentiate # a channel move and an actual force disconnect if channel_id is None: # We're being disconnected so cleanup await self.disconnect() else: guild = self.guild self.channel = channel_id and guild and guild.get_channel(int(channel_id)) # type: ignore else: self._voice_state_complete.set() async def on_voice_server_update(self, data: VoiceServerUpdatePayload) -> None: if self._voice_server_complete.is_set(): _log.info("Ignoring extraneous voice server update.") return self.token = data.get("token") self.server_id = int(data["guild_id"]) endpoint = data.get("endpoint") if endpoint is None or self.token is None: _log.warning( "Awaiting endpoint... This requires waiting. " "If timeout occurred considering raising the timeout and reconnecting." ) return self.endpoint, _, _ = endpoint.rpartition(":") if self.endpoint.startswith("wss://"): # Just in case, strip it off since we're going to add it later self.endpoint = self.endpoint[6:] # This gets set later self.endpoint_ip = MISSING self.socket = socket.socket(socket.AF_INET, socket.SOCK_DGRAM) self.socket.setblocking(False) if not self._handshaking: # If we're not handshaking then we need to terminate our previous connection in the websocket await self.ws.close(4000) return self._voice_server_complete.set() async def voice_connect(self) -> None: await self.channel.guild.change_voice_state(channel=self.channel) async def voice_disconnect(self) -> None: _log.info( "The voice handshake is being terminated for Channel ID %s (Guild ID %s)", self.channel.id, self.guild.id, ) await self.channel.guild.change_voice_state(channel=None) def prepare_handshake(self) -> None: self._voice_state_complete.clear() self._voice_server_complete.clear() self._handshaking = True _log.info( "Starting voice handshake... (connection attempt %d)", self._connections + 1 ) self._connections += 1 def finish_handshake(self) -> None: _log.info("Voice handshake complete. Endpoint found %s", self.endpoint) self._handshaking = False self._voice_server_complete.clear() self._voice_state_complete.clear() async def connect_websocket(self) -> DiscordVoiceWebSocket: ws = await DiscordVoiceWebSocket.from_client(self) self._connected.clear() while ws.secret_key is None: await ws.poll_event() self._connected.set() return ws async def connect(self, *, reconnect: bool, timeout: float) -> None: _log.info("Connecting to voice...") self.timeout = timeout for i in range(5): self.prepare_handshake() # This has to be created before we start the flow. futures = [ self._voice_state_complete.wait(), self._voice_server_complete.wait(), ] # Start the connection flow await self.voice_connect() try: await utils.sane_wait_for(futures, timeout=timeout) except asyncio.TimeoutError: await self.disconnect(force=True) raise self.finish_handshake() try: self.ws = await self.connect_websocket() break except (ConnectionClosed, asyncio.TimeoutError): if reconnect: _log.exception("Failed to connect to voice... Retrying...") await asyncio.sleep(1 + i * 2.0) await self.voice_disconnect() continue else: raise if self._runner is MISSING: self._runner = self.loop.create_task(self.poll_voice_ws(reconnect)) async def potential_reconnect(self) -> bool: # Attempt to stop the player thread from playing early self._connected.clear() self.prepare_handshake() self._potentially_reconnecting = True try: # We only care about VOICE_SERVER_UPDATE since VOICE_STATE_UPDATE can come before we get disconnected await asyncio.wait_for( self._voice_server_complete.wait(), timeout=self.timeout ) except asyncio.TimeoutError: self._potentially_reconnecting = False await self.disconnect(force=True) return False self.finish_handshake() self._potentially_reconnecting = False try: self.ws = await self.connect_websocket() except (ConnectionClosed, asyncio.TimeoutError): return False else: return True @property def latency(self) -> float: """Latency between a HEARTBEAT and a HEARTBEAT_ACK in seconds. This could be referred to as the Discord Voice WebSocket latency and is an analogue of user's voice latencies as seen in the Discord client. .. versionadded:: 1.4 """ ws = self.ws return float("inf") if not ws else ws.latency @property def average_latency(self) -> float: """Average of most recent 20 HEARTBEAT latencies in seconds. .. versionadded:: 1.4 """ ws = self.ws return float("inf") if not ws else ws.average_latency async def poll_voice_ws(self, reconnect: bool) -> None: backoff = ExponentialBackoff() while True: try: await self.ws.poll_event() except (ConnectionClosed, asyncio.TimeoutError) as exc: if isinstance(exc, ConnectionClosed): # The following close codes are undocumented, so I will document them here. # 1000 - normal closure (obviously) # 4014 - voice channel has been deleted. # 4015 - voice server has crashed if exc.code in (1000, 4015): _log.info( "Disconnecting from voice normally, close code %d.", exc.code, ) await self.disconnect() break if exc.code == 4014: _log.info( "Disconnected from voice by force... potentially" " reconnecting." ) successful = await self.potential_reconnect() if successful: continue _log.info( "Reconnect was unsuccessful, disconnecting from voice" " normally..." ) await self.disconnect() break if not reconnect: await self.disconnect() raise retry = backoff.delay() _log.exception( "Disconnected from voice... Reconnecting in %.2fs.", retry ) self._connected.clear() await asyncio.sleep(retry) await self.voice_disconnect() try: await self.connect(reconnect=True, timeout=self.timeout) except asyncio.TimeoutError: # at this point we've retried 5 times... let's continue the loop. _log.warning("Could not connect to voice... Retrying...") continue async def disconnect(self, *, force: bool = False) -> None: """|coro| Disconnects this voice client from voice. """ if not force and not self.is_connected(): return self.stop() self._connected.clear() try: if self.ws: await self.ws.close() await self.voice_disconnect() finally: self.cleanup() if self.socket: self.socket.close() async def move_to(self, channel: abc.Connectable) -> None: """|coro| Moves you to a different voice channel. Parameters ---------- channel: :class:`abc.Connectable` The channel to move to. Must be a voice channel. """ await self.channel.guild.change_voice_state(channel=channel) def is_connected(self) -> bool: """Indicates if the voice client is connected to voice.""" return self._connected.is_set() # audio related def _get_voice_packet(self, data): header = bytearray(12) # Formulate rtp header header[0] = 0x80 header[1] = 0x78 struct.pack_into(">H", header, 2, self.sequence) struct.pack_into(">I", header, 4, self.timestamp) struct.pack_into(">I", header, 8, self.ssrc) encrypt_packet = getattr(self, f"_encrypt_{self.mode}") return encrypt_packet(header, data) def _encrypt_xsalsa20_poly1305(self, header: bytes, data) -> bytes: box = nacl.secret.SecretBox(bytes(self.secret_key)) nonce = bytearray(24) nonce[:12] = header return header + box.encrypt(bytes(data), bytes(nonce)).ciphertext def _encrypt_xsalsa20_poly1305_suffix(self, header: bytes, data) -> bytes: box = nacl.secret.SecretBox(bytes(self.secret_key)) nonce = nacl.utils.random(nacl.secret.SecretBox.NONCE_SIZE) return header + box.encrypt(bytes(data), nonce).ciphertext + nonce def _encrypt_xsalsa20_poly1305_lite(self, header: bytes, data) -> bytes: box = nacl.secret.SecretBox(bytes(self.secret_key)) nonce = bytearray(24) nonce[:4] = struct.pack(">I", self._lite_nonce) self.checked_add("_lite_nonce", 1, 4294967295) return header + box.encrypt(bytes(data), bytes(nonce)).ciphertext + nonce[:4] def _decrypt_xsalsa20_poly1305(self, header, data): box = nacl.secret.SecretBox(bytes(self.secret_key)) nonce = bytearray(24) nonce[:12] = header return self.strip_header_ext(box.decrypt(bytes(data), bytes(nonce))) def _decrypt_xsalsa20_poly1305_suffix(self, header, data): box = nacl.secret.SecretBox(bytes(self.secret_key)) nonce_size = nacl.secret.SecretBox.NONCE_SIZE nonce = data[-nonce_size:] return self.strip_header_ext(box.decrypt(bytes(data[:-nonce_size]), nonce)) def _decrypt_xsalsa20_poly1305_lite(self, header, data): box = nacl.secret.SecretBox(bytes(self.secret_key)) nonce = bytearray(24) nonce[:4] = data[-4:] data = data[:-4] return self.strip_header_ext(box.decrypt(bytes(data), bytes(nonce))) @staticmethod def strip_header_ext(data): if data[0] == 0xBE and data[1] == 0xDE and len(data) > 4: _, length = struct.unpack_from(">HH", data) offset = 4 + length * 4 data = data[offset:] return data def get_ssrc(self, user_id): return {info["user_id"]: ssrc for ssrc, info in self.ws.ssrc_map.items()}[ user_id ] @overload def play( self, source: AudioSource, *, after: Callable[[Exception | None], Any] | None = None, wait_finish: Literal[False] = False, ) -> None: ... @overload def play( self, source: AudioSource, *, after: Callable[[Exception | None], Any] | None = None, wait_finish: Literal[True], ) -> asyncio.Future: ... def play( self, source: AudioSource, *, after: Callable[[Exception | None], Any] | None = None, wait_finish: bool = False, ) -> None | asyncio.Future: """Plays an :class:`AudioSource`. The finalizer, ``after`` is called after the source has been exhausted or an error occurred. If an error happens while the audio player is running, the exception is caught and the audio player is then stopped. If no after callback is passed, any caught exception will be displayed as if it were raised. Parameters ---------- source: :class:`AudioSource` The audio source we're reading from. after: Callable[[Optional[:class:`Exception`]], Any] The finalizer that is called after the stream is exhausted. This function must have a single parameter, ``error``, that denotes an optional exception that was raised during playing. wait_finish: bool If True, an awaitable will be returned, which can be used to wait for audio to stop playing. This awaitable will return an exception if raised, or None when no exception is raised. If False, None is returned and the function does not block. .. versionadded:: v2.5 Raises ------ ClientException Already playing audio or not connected. TypeError Source is not a :class:`AudioSource` or after is not a callable. OpusNotLoaded Source is not opus encoded and opus is not loaded. """ if not self.is_connected(): raise ClientException("Not connected to voice.") if self.is_playing(): raise ClientException("Already playing audio.") if not isinstance(source, AudioSource): raise TypeError( f"source must be an AudioSource not {source.__class__.__name__}" ) if not self.encoder and not source.is_opus(): self.encoder = opus.Encoder() future = None if wait_finish: future = asyncio.Future() after_callback = after def _after(exc: Exception | None): if callable(after_callback): after_callback(exc) future.set_result(exc) after = _after self._player = AudioPlayer(source, self, after=after) self._player.start() return future def unpack_audio(self, data): """Takes an audio packet received from Discord and decodes it into pcm audio data. If there are no users talking in the channel, `None` will be returned. You must be connected to receive audio. .. versionadded:: 2.0 Parameters ---------- data: :class:`bytes` Bytes received by Discord via the UDP connection used for sending and receiving voice data. """ if 200 <= data[1] <= 204: # RTCP received. # RTCP provides information about the connection # as opposed to actual audio data, so it's not # important at the moment. return if self.paused: return data = RawData(data, self) if data.decrypted_data == b"\xf8\xff\xfe": # Frame of silence return self.decoder.decode(data) def start_recording(self, sink, callback, *args, sync_start: bool = False): """The bot will begin recording audio from the current voice channel it is in. This function uses a thread so the current code line will not be stopped. Must be in a voice channel to use. Must not be already recording. .. versionadded:: 2.0 Parameters ---------- sink: :class:`.Sink` A Sink which will "store" all the audio data. callback: :ref:`coroutine ` A function which is called after the bot has stopped recording. *args: Args which will be passed to the callback function. sync_start: :class:`bool` If True, the recordings of subsequent users will start with silence. This is useful for recording audio just as it was heard. Raises ------ RecordingException Not connected to a voice channel. RecordingException Already recording. RecordingException Must provide a Sink object. """ if not self.is_connected(): raise RecordingException("Not connected to voice channel.") if self.recording: raise RecordingException("Already recording.") if not isinstance(sink, Sink): raise RecordingException("Must provide a Sink object.") self.empty_socket() self.decoder = opus.DecodeManager(self) self.decoder.start() self.recording = True self.sync_start = sync_start self.sink = sink sink.init(self) t = threading.Thread( target=self.recv_audio, args=( sink, callback, *args, ), ) t.start() def stop_recording(self): """Stops the recording. Must be already recording. .. versionadded:: 2.0 Raises ------ RecordingException Not currently recording. """ if not self.recording: raise RecordingException("Not currently recording audio.") self.decoder.stop() self.recording = False self.paused = False def toggle_pause(self): """Pauses or unpauses the recording. Must be already recording. .. versionadded:: 2.0 Raises ------ RecordingException Not currently recording. """ if not self.recording: raise RecordingException("Not currently recording audio.") self.paused = not self.paused def empty_socket(self): while True: ready, _, _ = select.select([self.socket], [], [], 0.0) if not ready: break for s in ready: s.recv(4096) def recv_audio(self, sink, callback, *args): # Gets data from _recv_audio and sorts # it by user, handles pcm files and # silence that should be added. self.user_timestamps: dict[int, tuple[int, float]] = {} self.starting_time = time.perf_counter() self.first_packet_timestamp: float while self.recording: ready, _, err = select.select([self.socket], [], [self.socket], 0.01) if not ready: if err: print(f"Socket error: {err}") continue try: data = self.socket.recv(4096) except OSError: self.stop_recording() continue self.unpack_audio(data) self.stopping_time = time.perf_counter() self.sink.cleanup() callback = asyncio.run_coroutine_threadsafe(callback(sink, *args), self.loop) result = callback.result() if result is not None: print(result) def recv_decoded_audio(self, data: RawData): # Add silence when they were not being recorded. if data.ssrc not in self.user_timestamps: # First packet from user if ( not self.user_timestamps or not self.sync_start ): # First packet from anyone self.first_packet_timestamp = data.receive_time silence = 0 else: # Previously received a packet from someone else silence = ( (data.receive_time - self.first_packet_timestamp) * 48000 ) - 960 else: # Previously received a packet from user dRT = ( data.receive_time - self.user_timestamps[data.ssrc][1] ) * 48000 # delta receive time dT = data.timestamp - self.user_timestamps[data.ssrc][0] # delta timestamp diff = abs(100 - dT * 100 / dRT) if ( diff > 60 and dT != 960 ): # If the difference in change is more than 60% threshold silence = dRT - 960 else: silence = dT - 960 self.user_timestamps.update({data.ssrc: (data.timestamp, data.receive_time)}) data.decoded_data = ( struct.pack(" bool: """Indicates if we're currently playing audio.""" return self._player is not None and self._player.is_playing() def is_paused(self) -> bool: """Indicates if we're playing audio, but if we're paused.""" return self._player is not None and self._player.is_paused() def stop(self) -> None: """Stops playing audio.""" if self._player: self._player.stop() self._player = None def pause(self) -> None: """Pauses the audio playing.""" if self._player: self._player.pause() def resume(self) -> None: """Resumes the audio playing.""" if self._player: self._player.resume() @property def source(self) -> AudioSource | None: """The audio source being played, if playing. This property can also be used to change the audio source currently being played. """ return self._player.source if self._player else None @source.setter def source(self, value: AudioSource) -> None: if not isinstance(value, AudioSource): raise TypeError(f"expected AudioSource not {value.__class__.__name__}.") if self._player is None: raise ValueError("Not playing anything.") self._player._set_source(value) def send_audio_packet(self, data: bytes, *, encode: bool = True) -> None: """Sends an audio packet composed of the data. You must be connected to play audio. Parameters ---------- data: :class:`bytes` The :term:`py:bytes-like object` denoting PCM or Opus voice data. encode: :class:`bool` Indicates if ``data`` should be encoded into Opus. Raises ------ ClientException You are not connected. opus.OpusError Encoding the data failed. """ self.checked_add("sequence", 1, 65535) if encode: if not self.encoder: self.encoder = opus.Encoder() encoded_data = self.encoder.encode(data, self.encoder.SAMPLES_PER_FRAME) else: encoded_data = data packet = self._get_voice_packet(encoded_data) try: self.socket.sendto(packet, (self.endpoint_ip, self.voice_port)) except BlockingIOError: _log.warning( "A packet has been dropped (seq: %s, timestamp: %s)", self.sequence, self.timestamp, ) self.checked_add("timestamp", opus.Encoder.SAMPLES_PER_FRAME, 4294967295)