make RTP timestamp strictly linear

this should fix some artificial delay

src/audio/manager.cpp

@@ -50,6 +50,15 @@ void capture_data_callback(ma_device *pDevice, void *pOutput, const void *pInput
     if (mgr == nullptr) return;
 
     mgr->OnCapturedPCM(static_cast<const int16_t *>(pInput), frameCount);
+
+    /*
+     * You can simply increment it by 480 in UDPSocket::SendEncrypted but this is wrong
+     * The timestamp is supposed to be strictly linear eg. if there's discontinuous
+     * transmission for 1 second then the timestamp should be 48000 greater than the
+     * last packet. So it's incremented here because this is fired 100x per second
+     * and is always called in sync with UDPSocket::SendEncrypted
+     */
+    mgr->m_rtp_timestamp += 480;
 }
 
 AudioManager::AudioManager() {
@@ -473,6 +482,10 @@ AudioDevices &AudioManager::GetDevices() {
     return m_devices;
 }
 
+uint32_t AudioManager::GetRTPTimestamp() const noexcept {
+    return m_rtp_timestamp;
+}
+
 AudioManager::type_signal_opus_packet AudioManager::signal_opus_packet() {
     return m_signal_opus_packet;
 }

src/audio/manager.hpp

@@ -63,6 +63,8 @@ public:
 
     [[nodiscard]] AudioDevices &GetDevices();
 
+    [[nodiscard]] uint32_t GetRTPTimestamp() const noexcept;
+
 private:
     void OnCapturedPCM(const int16_t *pcm, ma_uint32 frames);
 
@@ -113,6 +115,8 @@ private:
 
     AudioDevices m_devices;
 
+    std::atomic<uint32_t> m_rtp_timestamp = 0;
+
 public:
     using type_signal_opus_packet = sigc::signal<void(int payload_size)>;
     type_signal_opus_packet signal_opus_packet();

src/discord/voiceclient.cpp

@@ -49,17 +49,18 @@ void UDPSocket::SetSSRC(uint32_t ssrc) {
 
 void UDPSocket::SendEncrypted(const uint8_t *data, size_t len) {
     m_sequence++;
-    m_timestamp += 480; // this is important
+
+    const uint32_t timestamp = Abaddon::Get().GetAudio().GetRTPTimestamp();
 
     std::vector<uint8_t> rtp(12 + len + crypto_secretbox_MACBYTES, 0);
     rtp[0] = 0x80; // ver 2
     rtp[1] = 0x78; // payload type 0x78
     rtp[2] = (m_sequence >> 8) & 0xFF;
     rtp[3] = (m_sequence >> 0) & 0xFF;
-    rtp[4] = (m_timestamp >> 24) & 0xFF;
-    rtp[5] = (m_timestamp >> 16) & 0xFF;
-    rtp[6] = (m_timestamp >> 8) & 0xFF;
-    rtp[7] = (m_timestamp >> 0) & 0xFF;
+    rtp[4] = (timestamp >> 24) & 0xFF;
+    rtp[5] = (timestamp >> 16) & 0xFF;
+    rtp[6] = (timestamp >> 8) & 0xFF;
+    rtp[7] = (timestamp >> 0) & 0xFF;
     rtp[8] = (m_ssrc >> 24) & 0xFF;
     rtp[9] = (m_ssrc >> 16) & 0xFF;
     rtp[10] = (m_ssrc >> 8) & 0xFF;

src/discord/voiceclient.hpp

@@ -171,7 +171,6 @@ private:
     uint32_t m_ssrc;
 
     uint16_t m_sequence = 0;
-    uint32_t m_timestamp = 0;
 
 public:
     using type_signal_data = sigc::signal<void, std::vector<uint8_t>>;