WebRTC Microphone Button Not Appearing for go2rtc Exec Stream with Backchannel

[woeiwen]

I'm trying to digitalize my Fermax 4+N analog intercom system with cvbs video using a Raspberry Pi 4B with Home Assistant OS and the go2rtc addon. I've successfully got video and audio working in the forward direction (intercom → browser), but I can't get the microphone button to appear in the WebRTC card for two-way audio.
Hardware Setup:

Fermax 4+N intercom (CVBS video + analog audio)
EasyCap USB capture device (video + audio input)
RPi4B audio output via PulseAudio to intercom speaker
go2rtc version: 1.9.14
WebRTC Camera card version: 3.6.1

Current go2rtc Configuration:

yamlstreams:
  intercom:
    - "exec:ffmpeg -f v4l2 -input_format yuyv422 -video_size 320x240 -i /dev/video0 -f pulse -i alsa_input.usb-MACROSILICON_AV_to_USB2.0_20200909-02.analog-stereo -c:v h264_v4l2m2m -b:v 500k -g 10 -c:a libopus -ar 16000 -ac 1 -b:a 10k -f rtsp {output}"
    - "exec:ffmpeg -fflags nobuffer -f s16le -ar 16000 -ac 1 -i pipe: -c:a pcm_s16le -ar 48000 -ac 2 -af 'aresample=async=1' -f pulse alsa_output.platform-fe00b840.mailbox.stereo-fallback#backchannel=1"

webrtc:
  candidates:
    - 192.168.2.199:8555
    - stun:8555
  ice_servers:
    - urls: [stun:stun.l.google.com:19302]

api:
  origin: '*'

Lovelace Card:

yamltype: custom:webrtc-camera
url: intercom
mode: webrtc
media: video,audio,microphone
ui: true

What Works:
✅ Forward video and audio (intercom → browser) works perfectly
✅ Backchannel audio pipeline works when I stream web radio or media_player content to it
✅ The second exec line successfully outputs audio to the intercom speaker
What Doesn't Work:
❌ Microphone button doesn't appear in the WebRTC card UI
❌ Microphone button doesn't appear in the go2rtc web player (http://ip:1984/)
What I've Verified:

The go2rtc API shows the stream has "audio, recvonly, ANY" in the consumer medias
There's a pipe producer created ("protocol": "pipe", "medias": ["audio, sendonly, PCML/16000"])
The routing exists: probe consumer receiver → pipe producer sender

The Problem:
Even though the #backchannel=1 tag creates the audio pipeline infrastructure, go2rtc doesn't seem to advertise bidirectional audio capability to WebRTC clients. The stream appears as send-only (video/audio out) rather than bidirectional.
Similar Working Projects:
I found this ESPHome intercom project that successfully implements two-way audio: https://github.com/n-IA-hane/esphome-intercom
Their setup uses UDP streams for bidirectional audio, but the key difference is they have separate audio-only streams, not video+audio combined like my use case.

Questions:

1. Is #backchannel=1 supposed to work with WebRTC exec streams in go2rtc 1.9.14?
2. Do I need a different approach to make the stream advertise bidirectional capability?
3. Has anyone successfully implemented WebRTC two-way audio with go2rtc using exec streams with UVC cameras or USB capture cards (not native two-way RTSP/ONVIF cameras with built-in backchannel support)?
4. Should I be using a different stream configuration pattern?

Thank you

[AlexxIT]

You should start from this part of documentation https://github.com/AlexxIT/go2rtc?tab=readme-ov-file#two-way-audio

[woeiwen]

First of all. Thank you Alex for this great project with amazing features and prompt reply. I have relooked at the docs which you pointed out. Spent some time, still trying to find out what is amiss in my setup as a novice. Could you give a hint ? Much thanks.

[AlexxIT]

Quote from the documentation:

The browser will give access to the microphone only for HTTPS sites

[woeiwen]

Much thanks Alex, Noted. I have been using https:// throughout. Have use the dashboard link to even start a browser as camera session on a one machine and to consume it on another. The mic works fine even. So browser mic works, I see that opus data stream (the net diagram attached

)is passed over to the server (192.168.2.199) but does not get piped to the pulse audio sink. Maybe the format mismatch on the backchannel stream.

[AlexxIT]

OK. It turns out your question is not about the microphone button, but about the operation of the exec backchannel.
The idea is that go2rtc doesn't know in advance which codec your exec backchannel supports. You need to specify it manually.
And of course, your audio source must output audio in this codec.

You think exec will receive audio in PCML/16000/1 format. But the browser cannot output sound in this format.

ffmpeg -fflags nobuffer -f s16le -ar 16000 -ac 1 -i pipe:

https://github.com/AlexxIT/go2rtc/tree/master/internal/exec#backchannel

[woeiwen]

a big thank you to you, alex. The browser can only do opus/48k, pcma and pcmu. my bad didn't read the docs properly.
it now works. with pcma/8k. but there is significant latency due to conversion. can't get opus to work though.

intercom:
- "exec:ffmpeg -f v4l2 -input_format yuyv422 -video_size 320x240 -i /dev/video0 -f pulse -i alsa_input.usb-MACROSILICON_AV_to_USB2.0_20200909-02.analog-stereo -c:v h264_v4l2m2m -b:v 500k -g 15 -c:a libopus -ar 48000 -ac 2 -b:a 64k -f rtsp {output}"
- "exec:ffmpeg -hide_banner -v error -fflags nobuffer+flush_packets -flags low_delay -f alaw -ar 8000 -ac 1 -i pipe:0 -f pulse alsa_output.platform-fe00b840.mailbox.stereo-fallback#backchannel=1#audio=alaw/8000"
webrtc-go2rtc: webrtc:ws://192.168.2.199:1984/api/ws?src=intercom

[olivluca]

Same problem with opus.

If I use, e.g.

- exec:pw-play --target=sink.echo-cancel-ref --quality=15 --format=alaw --rate=8000 --channels=1 --latency=64ms --raw -#backchannel=1#audio=alaw/8000

I get audio from the browser, it plays, but it is horrible.

If I use
- exec:/home/luca/backchannel.sh#backchannel=1#audio=opus/48000/2

the browser sends audio but the data consist of rdp packets that I don't know how to decode and play (to test I just used cat in backchannel.sh and I found no way to decode the resulting file).