j.foucher/PS_AI_Agent
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

ConvAgent: fire the voice-onset interrupt server-side (dedicated-server support)

Browse Source 
The agent-to-agent dialogue interrupt was detected only in FeedExternalAudio,
which runs client-side for a remote player. On a DEDICATED server (no host
player) no one could trigger the reliable VAD interrupt -- everyone fell back to
the slower ElevenLabs transcript path.

Also broadcast OnUserVoiceDetected from ServerSendMicAudioFromPlayer, where the
player's mic (the host's OR a remote client's, relayed) always arrives on the
server. Shares the LastUserVoiceOnset debounce so the host does not double-fire.
The orchestrator still attributes the correct player (ResolveSpeakingPlayer
finds the connected client via NetConnectedPawns).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
j.foucher
2026-07-02 17:18:10 +02:00
parent f49a57f892
commit 087c83018c
1 changed files with 26 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
Unreal/PS_AI_Agent/Plugins/PS_AI_ConvAgent/Source/PS_AI_ConvAgent/Private/PS_AI_ConvAgent_ElevenLabsComponent.cpp
View File

@@ -2525,6 +2525,32 @@ void UPS_AI_ConvAgent_ElevenLabsComponent::ServerSendMicAudioFromPlayer(
{
if (!WebSocketProxy || !WebSocketProxy->IsConnected()) return;
// Server-side voice-onset. This runs on the SERVER (only the server has a live
// WebSocket — clients returned above), and the player's mic — the host's OR a
// remote client's, relayed here — always passes through this function. Broadcasting
// OnUserVoiceDetected here is what makes the agent-to-agent dialogue interrupt work
// on a DEDICATED server (no host player). Debounced (shares LastUserVoiceOnset with
// the FeedExternalAudio path, so the host doesn't double-fire).
{
const int32 OnsetSamples = PCMBytes.Num() / (int32)sizeof(int16);
if (OnsetSamples > 0)
{
const int16* OS = reinterpret_cast<const int16*>(PCMBytes.GetData());
double OnsetSumSq = 0.0;
for (int32 i = 0; i < OnsetSamples; ++i) { const double v = OS[i] / 32768.0; OnsetSumSq += v * v; }
const double OnsetRms = FMath::Sqrt(OnsetSumSq / OnsetSamples);
if (OnsetRms >= 0.02)
{
const double NowSec = FPlatformTime::Seconds();
if (NowSec - LastUserVoiceOnset > 0.5)
{
LastUserVoiceOnset = NowSec;
OnUserVoiceDetected.Broadcast();
}
}
}
}
// Standalone / single-player: bypass speaker arbitration entirely.
// There's only one player — no need for Contains check or speaker switching.
if (NetConnectedPawns.Num() <= 1)