PLC generating silence? about opus

[jmvalin]

You're seeing pure silence for the missing frames or are you just saying there are artefacts? Also, keep in mind that the PLC tends to work better for speech than music. That being said, it should always do something and not just insert silence.

[ArnaudBienner]

@jmvalin I'm seeing pure silence.
I tested with a movie trailer, which contain both music and speech.
If you believe it should work on this case too, I will try to get a short extract to share with you, if you think that can help?

[jmvalin]

Yes, please share the following:

1. the input audio file
2. the exact command line you're using
3. the output you're seeing

[ArnaudBienner]

Hi,

I've put the files here (source and generated by libopus):
https://drive.google.com/open?id=1Ac5sPuzCLWsEcen8UILFtU7jXykLSPIB

I've generated the .wav from a Blender movie 1 2 using the following command:

ffmpeg -i ToS-4k-1920.mov  ToS-4k-1920.wav

Then, with libopus 1.2.1 I've compiled locally on my macOS 10.11:

.libs/opus_demo audio 48000 2 128000 ToS-4k-1920.wav ToS_noloss.raw
.libs/opus_demo audio 48000 2 128000 -loss 1 ToS-4k-1920.wav ToS_loss1.raw

I've opened the generated files in Audacity using the following settings:
Encoding: Signed 16 bits
Little-endian
Stereo
Sample Rate: 44100 Hz; Note: I'm not sure why output is 44100: this was the source sample rate, but I specified 48kHz in the opus_demo command line

Listening to the audio with 1 percent loss, I can regularly hear some silences.
And they are visible in the Audicity, as you can see in this screenshot:

I had the same problem with another wav file, with sample rate 48kHz.

What do you think?

[jmvalin]

Can you check if master still has the problem? There was a PLC bug in 1.2

[ArnaudBienner]

That seems better: I can hear some artifact (not always) but at least it's not just silence.
I guess PLC can't compensate for all packets loss.