Legacy mute causes audio session switch

[dfed]

Describe the bug
Muting the local participant's microphone when isLegacyMuteMode is enabled mutes all audio, including agent audio and other sound effects being played by AVAudioEngine

SDK Version
2.2.1. This is a regression from 2.0.19. I have not tested releases between these.

iOS/macOS Version
Reproduced on iOS 17.7.2 and iOS 18.4b1

Xcode Version
16.2.0 (Swift 6 mode)

Steps to Reproduce

1. AudioManager.shared.isLegacyMuteMode = true
2. Connect to a room.
3. Call room.localParticipant.setMicrophone(enabled: false)

Expected behavior
Microphone input is muted, and other audio is not muted.

[hiroshihorie]

Hi @dfed I'll investigate this today. I've already tested this case so I wonder what's going on 🤔

[sergeyphi]

Hi @hiroshihorie I've also noticed this issue.

I've also noticed that the muted state will have it's own volume level that the app will "remember".

I'll expand @dfed's repro steps to show what I mean:

1. AudioManager.shared.isLegacyMuteMode = true
2. Connect to a room
3. Change volume to 1.0 with iPhone buttons
4. Call room.localParticipant.setMicrophone(enabled: false)
5. Observe not being able remote participant's audio track (potentially)
6. Change volume to 0.5 with iPhone buttons
7. Now you can hear remote participant
8. room.localParticipant.setMicrophone(enabled: true)
9. Observe audio volume changed back to 1.0
10. room.localParticipant.setMicrophone(enabled: false)
11. Observe audio volume changed to 0.5 (value you set before while muted)

Sharing in case it's useful!

[dfed]

Fascinating observation! I believe that would indicate we’re changing the the audio session configuration when disabling the microphone.

[hiroshihorie]

Thank you, @dfed and @sergeyphi. Yes, the audio session switch is causing the volume level change. When legacyMuteMode is enabled, muting shuts down the recording, which then switches to .playback.

I’ll implement a fix or provide an option to modify this behavior. In a mute/unmute scenario you probably just want to keep it on .playAndRecord. 🙂

[hiroshihorie]

Hi, I'm working on fix in #621 but I've discovered the following.

When maintaining category: .playAndRecord mode: .videoChat and disabling recording,
volume slider remains same which is good ✅ but now the output volume drops significantly ❌

Using this mode without the VoiceProcessing IO unit or AVAudioEngine with voice processing enabled will result in the following:
    - Chat-specific signal processing such as echo cancellation or automatic gain correction will not be loaded
    - Dynamic processing on input and output will be disabled resulting in a lower output playback level. 

If I switch to another mode, even with same category (.playAndRecord mode: .default),
volume slider switches ❌ but the output volume is good ✅

Example:
https://github.com/user-attachments/assets/aee8e81e-2c0a-4012-a871-948be8033879

So I'm a little stuck at the moment.

Another option is to keep the recording on (mic indicator on) but disregard local audio buffers (mute) which I'm sure will work.
But I assume turning off the mic indicator is a requirement ?

[dfed]

Ahhhhhh right I remember hitting this before we adopted LiveKit – turning off voice processing enables you to turn off the audio indicator but it also makes all audio playback quite quiet.

Dropping output volume is a no-go, since there's no way for the user to work around it. Switching modes changes the volume, which is not great, but at least the user can fix it + develop a mental model around it. I think this is the preferred approach.

Another option is to keep the recording on (mic indicator on) but disregard local audio buffers (mute) which I'm sure will work.

I agree this'll work, but I also agree that this doesn't feel viable.

Edit: what I don't understand is how the modern mute mode avoids this pitfall. What's the setting in the audio stack that enables us to have both consistent volume across mute/unmute as well as voice processing?

[hiroshihorie]

The "modern mute mode" is using isVoiceProcessingInputMuted which is exactly for this purpose. But unfortunately plays a sound effect, probably for privacy reasons. 🤔