| /* |
| * Copyright 2018 The WebRTC project authors. All Rights Reserved. |
| * |
| * Use of this source code is governed by a BSD-style license |
| * that can be found in the LICENSE file in the root of the source |
| * tree. An additional intellectual property rights grant can be found |
| * in the file PATENTS. All contributing project authors may |
| * be found in the AUTHORS file in the root of the source tree. |
| */ |
| |
| // This is EXPERIMENTAL interface for media transport. |
| // |
| // The goal is to refactor WebRTC code so that audio and video frames |
| // are sent / received through the media transport interface. This will |
| // enable different media transport implementations, including QUIC-based |
| // media transport. |
| |
| #ifndef API_MEDIA_TRANSPORT_INTERFACE_H_ |
| #define API_MEDIA_TRANSPORT_INTERFACE_H_ |
| |
| #include <memory> |
| #include <utility> |
| #include <vector> |
| |
| #include "api/rtcerror.h" |
| #include "common_types.h" // NOLINT(build/include) |
| |
| namespace rtc { |
| class PacketTransportInternal; |
| class Thread; |
| } // namespace rtc |
| |
| namespace webrtc { |
| |
| // Represents encoded audio frame in any encoding (type of encoding is opaque). |
| // To avoid copying of encoded data use move semantics when passing by value. |
| class MediaTransportEncodedAudioFrame { |
| public: |
| enum class FrameType { |
| // Normal audio frame (equivalent to webrtc::kAudioFrameSpeech). |
| kSpeech, |
| |
| // DTX frame (equivalent to webrtc::kAudioFrameCN). |
| kDiscountinuousTransmission, |
| }; |
| |
| MediaTransportEncodedAudioFrame( |
| // Audio sampling rate, for example 48000. |
| int sampling_rate_hz, |
| |
| // Starting sample index of the frame, i.e. how many audio samples were |
| // before this frame since the beginning of the call or beginning of time |
| // in one channel (the starting point should not matter for NetEq). In |
| // WebRTC it is used as a timestamp of the frame. |
| // TODO(sukhanov): Starting_sample_index is currently adjusted on the |
| // receiver side in RTP path. Non-RTP implementations should preserve it. |
| // For NetEq initial offset should not matter so we should consider fixing |
| // RTP path. |
| int starting_sample_index, |
| |
| // Number of audio samples in audio frame in 1 channel. |
| int samples_per_channel, |
| |
| // Sequence number of the frame in the order sent, it is currently |
| // required by NetEq, but we can fix NetEq, because starting_sample_index |
| // should be enough. |
| int sequence_number, |
| |
| // If audio frame is a speech or discontinued transmission. |
| FrameType frame_type, |
| |
| // Opaque payload type. In RTP codepath payload type is stored in RTP |
| // header. In other implementations it should be simply passed through the |
| // wire -- it's needed for decoder. |
| uint8_t payload_type, |
| |
| // Vector with opaque encoded data. |
| std::vector<uint8_t> encoded_data) |
| : sampling_rate_hz_(sampling_rate_hz), |
| starting_sample_index_(starting_sample_index), |
| samples_per_channel_(samples_per_channel), |
| sequence_number_(sequence_number), |
| frame_type_(frame_type), |
| payload_type_(payload_type), |
| encoded_data_(std::move(encoded_data)) {} |
| |
| // Getters. |
| int sampling_rate_hz() const { return sampling_rate_hz_; } |
| int starting_sample_index() const { return starting_sample_index_; } |
| int samples_per_channel() const { return samples_per_channel_; } |
| int sequence_number() const { return sequence_number_; } |
| |
| uint8_t payload_type() const { return payload_type_; } |
| FrameType frame_type() const { return frame_type_; } |
| |
| rtc::ArrayView<const uint8_t> encoded_data() const { return encoded_data_; } |
| |
| private: |
| int sampling_rate_hz_; |
| int starting_sample_index_; |
| int samples_per_channel_; |
| |
| // TODO(sukhanov): Refactor NetEq so we don't need sequence number. |
| // Having sample_index and samples_per_channel should be enough. |
| int sequence_number_; |
| |
| FrameType frame_type_; |
| |
| // TODO(sukhanov): Consider enumerating allowed encodings and store enum |
| // instead of uint payload_type. |
| uint8_t payload_type_; |
| |
| std::vector<uint8_t> encoded_data_; |
| }; |
| |
| // Interface for receiving encoded audio frames from MediaTransportInterface |
| // implementations. |
| class MediaTransportAudioSinkInterface { |
| public: |
| virtual ~MediaTransportAudioSinkInterface() = default; |
| |
| // Called when new encoded audio frame is received. |
| virtual void OnData(uint64_t channel_id, |
| MediaTransportEncodedAudioFrame frame) = 0; |
| }; |
| |
| // Represents encoded video frame, along with the codec information. |
| class MediaTransportEncodedVideoFrame { |
| public: |
| MediaTransportEncodedVideoFrame(int64_t frame_id, |
| std::vector<int64_t> referenced_frame_ids, |
| VideoCodecType codec_type, |
| const webrtc::EncodedImage& encoded_image) |
| : codec_type_(codec_type), |
| encoded_image_(encoded_image), |
| frame_id_(frame_id), |
| referenced_frame_ids_(std::move(referenced_frame_ids)) {} |
| |
| VideoCodecType codec_type() const { return codec_type_; } |
| const webrtc::EncodedImage& encoded_image() const { return encoded_image_; } |
| |
| int64_t frame_id() const { return frame_id_; } |
| const std::vector<int64_t>& referenced_frame_ids() const { |
| return referenced_frame_ids_; |
| } |
| |
| private: |
| VideoCodecType codec_type_; |
| |
| // The buffer is not owned by the encoded image by default. On the sender it |
| // means that it will need to make a copy of it if it wants to deliver it |
| // asynchronously. |
| webrtc::EncodedImage encoded_image_; |
| |
| // Frame id uniquely identifies a frame in a stream. It needs to be unique in |
| // a given time window (i.e. technically unique identifier for the lifetime of |
| // the connection is not needed, but you need to guarantee that remote side |
| // got rid of the previous frame_id if you plan to reuse it). |
| // |
| // It is required by a remote jitter buffer, and is the same as |
| // EncodedFrame::id::picture_id. |
| // |
| // This data must be opaque to the media transport, and media transport should |
| // itself not make any assumptions about what it is and its uniqueness. |
| int64_t frame_id_; |
| |
| // A single frame might depend on other frames. This is set of identifiers on |
| // which the current frame depends. |
| std::vector<int64_t> referenced_frame_ids_; |
| }; |
| |
| // Interface for receiving encoded video frames from MediaTransportInterface |
| // implementations. |
| class MediaTransportVideoSinkInterface { |
| public: |
| virtual ~MediaTransportVideoSinkInterface() = default; |
| |
| // Called when new encoded video frame is received. |
| virtual void OnData(uint64_t channel_id, |
| MediaTransportEncodedVideoFrame frame) = 0; |
| |
| // Called when the request for keyframe is received. |
| virtual void OnKeyFrameRequested(uint64_t channel_id) = 0; |
| }; |
| |
| // Media transport interface for sending / receiving encoded audio/video frames |
| // and receiving bandwidth estimate update from congestion control. |
| class MediaTransportInterface { |
| public: |
| virtual ~MediaTransportInterface() = default; |
| |
| // Start asynchronous send of audio frame. The status returned by this method |
| // only pertains to the synchronous operations (e.g. |
| // serialization/packetization), not to the asynchronous operation. |
| |
| virtual RTCError SendAudioFrame(uint64_t channel_id, |
| MediaTransportEncodedAudioFrame frame) = 0; |
| |
| // Start asynchronous send of video frame. The status returned by this method |
| // only pertains to the synchronous operations (e.g. |
| // serialization/packetization), not to the asynchronous operation. |
| virtual RTCError SendVideoFrame( |
| uint64_t channel_id, |
| const MediaTransportEncodedVideoFrame& frame) = 0; |
| |
| // Requests a keyframe for the particular channel (stream). The caller should |
| // check that the keyframe is not present in a jitter buffer already (i.e. |
| // don't request a keyframe if there is one that you will get from the jitter |
| // buffer in a moment). |
| virtual RTCError RequestKeyFrame(uint64_t channel_id) = 0; |
| |
| // Sets audio sink. Sink must be unset by calling SetReceiveAudioSink(nullptr) |
| // before the media transport is destroyed or before new sink is set. |
| virtual void SetReceiveAudioSink(MediaTransportAudioSinkInterface* sink) = 0; |
| |
| // Registers a video sink. Before destruction of media transport, you must |
| // pass a nullptr. |
| virtual void SetReceiveVideoSink(MediaTransportVideoSinkInterface* sink) = 0; |
| |
| // TODO(sukhanov): RtcEventLogs. |
| // TODO(sukhanov): Bandwidth updates. |
| }; |
| |
| // If media transport factory is set in peer connection factory, it will be |
| // used to create media transport for sending/receiving encoded frames and |
| // this transport will be used instead of default RTP/SRTP transport. |
| // |
| // Currently Media Transport negotiation is not supported in SDP. |
| // If application is using media transport, it must negotiate it before |
| // setting media transport factory in peer connection. |
| class MediaTransportFactory { |
| public: |
| virtual ~MediaTransportFactory() = default; |
| |
| // Creates media transport. |
| // - Does not take ownership of packet_transport or network_thread. |
| // - Does not support group calls, in 1:1 call one side must set |
| // is_caller = true and another is_caller = false. |
| virtual RTCErrorOr<std::unique_ptr<MediaTransportInterface>> |
| CreateMediaTransport(rtc::PacketTransportInternal* packet_transport, |
| rtc::Thread* network_thread, |
| bool is_caller) = 0; |
| }; |
| |
| } // namespace webrtc |
| #endif // API_MEDIA_TRANSPORT_INTERFACE_H_ |