0
votes

I have this pipeline that uses pocketsphinx's VAD element :

            Gst.parse_launch(
              "pulsesrc device=\"alsa_input.usb-046d_08c9_674634A4-02-U0x46d0x8c9.analog-mono\" " +
              "! vader name=vad auto-threshold=true " +
              "! level name=wavelevel interval=100000000 " + // level interval is in nanoseconds 
              "! wavenc " +
              "! filesink location=audioz.wav"
            );

It works fine except that the streaming stops when there is no voice comming in the source.

I want to recording to continue independently of the VAD, so I tried this pipeline with a tee :

            Gst.parse_launch(
              "pulsesrc device=\"alsa_input.usb-046d_08c9_674634A4-02-U0x46d0x8c9.analog-mono\" " +
              "! tee name=t " +
              "! queue " +
              "! vader name=vad auto-threshold=true " +
              "! fakesink t. sync=false" +
              "! queue " +
              "! level name=wavelevel interval=100000000 " + // level interval is in nanoseconds 
              "! wavenc " +
              "! filesink location=audioz.wav"
            );

And this one is always stalling, state going from NULL -> READY -> PAUSE, and stalling forever on PAUSE.

The goal of the "independent VAD" is simply to record the begining and end time of voice segments (detected by the VAD).

Update :

commenting the line : "! fakesink t. sync=false" solves the problem, so the following pipeline does what I need :

            this.pipeline = Gst.parse_launch(
              "pulsesrc device=\"alsa_input.usb-046d_08c9_674634A4-02-U0x46d0x8c9.analog-mono\" " +
              " ! tee name=t" +
              " t. ! queue " +
              " ! vader name=vad auto-threshold=true " +              
              " t. ! queue " +
              " ! level name=wavelevel interval=1000000000 " + // level interval is in nanoseconds 
              " ! wavenc " +
              " ! filesink location=audioz.wav"
            );

The remaining question is if it's OK to have a queue without a sink...

1
It's probably easier to modify the vader itself (add property to it) to bypass the audio even if it's silence.Nikolay Shmyrev

1 Answers

1
votes

The easiest solution is to set async=0 on both sinks. (fakesink and filesink)