View Source Membrane.RTP.VAD (Membrane RTP plugin v0.21.0)

Simple vad based on audio level sent in RTP header.

To make this module work appropriate RTP header extension has to be set in SDP offer/answer.

If avg of audio level in packets in time_window exceeds vad_threshold it emits Membrane.RTP.VadEvent on its output pad.

When avg falls below vad_threshold and doesn't exceed it in the next vad_silence_timer it also emits the event.

Buffers that are processed by this element may or may not have been processed by a depayloader and passed through a jitter buffer. If they have not, then the only timestamp available for time comparison is the RTP timestamp. The delta between RTP timestamps is dependent on the clock rate used by the encoding. For OPUS the clock rate is 48kHz and packets are sent every 20ms, so the RTP timestamp delta between sequential packets should be 48000 / 1000 * 20, or 960.

When calculating the epoch of the timestamp, we need to account for 32bit integer wrapping.

  • :current - the difference between timestamps is low: the timestamp has not wrapped around.
  • :next - the timestamp has wrapped around to 0. To simplify queue processing we reset the state.
  • :prev - the timestamp has recently wrapped around. We might receive an out-of-order packet from before the rollover, which we ignore.


Element options

Passed via struct Membrane.RTP.VAD.t/0

  • vad_id


    ID of VAD header extension.

  • clock_rate


    Default value: 48000
    Clock rate (in Hz) for the encoding.

  • time_window


    Default value: 2000
    Time window (in ms) in which avg audio level is measured.

  • min_packet_num


    Default value: 50
    Minimal number of packets to count avg audio level from. Speech won't be detected until there are enough packets.

  • vad_threshold


    Default value: -50
    Audio level in dBov representing vad threshold. Values above are considered to represent voice activity. Value -127 represents digital silence.

  • vad_silence_time


    Default value: 300
    Time to wait before emitting Membrane.RTP.VadEvent after audio track is no longer considered to represent speech. If at this time audio track is considered to represent speech again the event will not be sent.





Accepted formats:

Demand mode::auto
Demand unit::buffers



Accepted formats:

Demand mode::auto

Link to this section Summary



Struct containing options for Membrane.RTP.VAD


Returns description of options available for this module

Link to this section Types

@type t() :: %Membrane.RTP.VAD{
  clock_rate: Membrane.RTP.clock_rate_t(),
  min_packet_num: pos_integer(),
  time_window: pos_integer(),
  vad_id: 1..14,
  vad_silence_time: pos_integer(),
  vad_threshold: -127..0

Struct containing options for Membrane.RTP.VAD

Link to this section Functions

@spec options() :: keyword()

Returns description of options available for this module