GNOME Bugzilla – Bug 779138
webrtcdsp: expose voice activity detection.
Last modified: 2017-06-02 19:35:45 UTC
The element now exposes properties to enable and configure voice activity detection, and posts "voice-activity" messages when the return value of stream_has_voice () changes.
Created attachment 346580 [details] [review] webrtcdsp: expose voice activity detection.
Review of attachment 346580 [details] [review]: Looks good to me. Just have a small question below, nothing important really. One note, if the VAD buffer size it's not a multiple of 10ms, our message timestamp will be slightly off. Maybe we should document ? Otherwise it's good for merging. ::: ext/webrtcdsp/gstwebrtcdsp.cpp @@ +624,3 @@ + if (self->voice_detection) { + GEnumClass *likelihood_class = (GEnumClass *) + g_type_class_ref (GST_TYPE_WEBRTC_VOICE_DETECTION_LIKELIHOOD); Considering this type is used for a property, I'm wondering if using g_type_class_peek() could be sufficient (mostly a curiosity).
Some local testing showed that it does not work if voice-detection-frame-size-ms is bigger of equal to 20ms. Do you know why ? Shall we document ?
Comment on attachment 346580 [details] [review] webrtcdsp: expose voice activity detection. Leaving open so we can investigate if it's a library bug or our bug. Attachment 346580 [details] pushed as fc261bc - webrtcdsp: expose voice activity detection.
Let's assume it's library limitation.