Question about voice-activity-detection

[lewiszlw]

Hi team,

Is this project support vad? I didn't find any vad example. Search code and only one place contains

/// AnswerOptions structure describes the options used to control the answer
/// creation process.
#[derive(Default, Debug, PartialEq, Eq, Copy, Clone)]
pub struct RTCAnswerOptions {
    /// voice_activity_detection allows the application to provide information
    /// about whether it wishes voice detection feature to be enabled or disabled.
    pub voice_activity_detection: bool,
}

/// OfferOptions structure describes the options used to control the offer
/// creation process
#[derive(Default, Debug, PartialEq, Eq, Copy, Clone)]
pub struct RTCOfferOptions {
    /// voice_activity_detection allows the application to provide information
    /// about whether it wishes voice detection feature to be enabled or disabled.
    pub voice_activity_detection: bool,

    /// ice_restart forces the underlying ice gathering process to be restarted.
    /// When this value is true, the generated description will have ICE
    /// credentials that are different from the current credentials
    pub ice_restart: bool,
}

If supports, could you share some doc/example I can learn?

[k0nserv]

Hey, this flag seems to have been obsoleted by the specification and doesn't do anything currently.

A way to do voice activity detection is to add the audio level extension to the negotiated headers

    media_engine
        .register_header_extension(
            webrtc::RTCRtpHeaderExtensionCapability {
                uri: String::from("urn:ietf:params:rtp-hdrext:ssrc-audio-level"),
            },
            webrtc::RTPCodecType::Audio,
            None,
        )

With this the other side will send audio level data in all audio RTP packets that you can extract to detect silence vs audio. In browsers you can use RTCRtpReceiver.getContributingSources()[0].audioLevel to get the audio level for an instant in time.

[lewiszlw]

Thanks for the quick answer. Sorry I find my needs are still unclear somewhere. Resolving first.