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

ConvAgent: add waveform-splice discontinuity debug detector

Browse Source 
Diagnostic for the audio "pops": EnqueueAgentAudio compares the last sample of
the previous enqueued block with the first sample of the new one and logs
[AUDIO-DBG] with the delta + a running count when the jump exceeds a threshold
(gated behind ps.ai.ConvAgent.Debug.ElevenLabs; role-tagged). The host's raw-PCM
stream is continuous so its count stays ~0; a client whose count climbs at
message cadence still has a splice problem. Confirms the carry-over fix
(bcb6c44) at runtime.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
j.foucher
2026-07-01 16:45:16 +02:00
parent bcb6c44f57
commit 6d92ec3c26
2 changed files with 45 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

37
Unreal/PS_AI_Agent/Plugins/PS_AI_ConvAgent/Source/PS_AI_ConvAgent/Private/PS_AI_ConvAgent_ElevenLabsComponent.cpp
View File

@@ -1732,6 +1732,42 @@ void UPS_AI_ConvAgent_ElevenLabsComponent::OnProceduralUnderflow(
void UPS_AI_ConvAgent_ElevenLabsComponent::EnqueueAgentAudio(const TArray<uint8>& PCMData)
{
// Debug: flag a waveform discontinuity at the splice between the previous
// enqueued block and this one — the signature of the audio pops. The host's
// raw-PCM stream is continuous so its count should stay ~0; if a client's
// count climbs at message cadence, a splice problem remains.
{
const int32 DbgVal = CVarDebugElevenLabs.GetValueOnGameThread();
const bool bAudioDbg = (DbgVal >= 0) ? (DbgVal > 0) : bDebug;
if (bAudioDbg && PCMData.Num() >= static_cast<int32>(sizeof(int16)))
{
const int16* Samples = reinterpret_cast<const int16*>(PCMData.GetData());
const int32 NumSamples = PCMData.Num() / static_cast<int32>(sizeof(int16));
if (bHasLastEnqueuedSample)
{
const int32 Delta = FMath::Abs(static_cast<int32>(Samples[0]) - static_cast<int32>(LastEnqueuedSample));
static constexpr int32 DiscontinuityThreshold = 3000; // |delta| on the int16 scale
if (Delta > DiscontinuityThreshold)
{
++AudioSpliceDiscontinuityCount;
UE_LOG(LogPS_AI_ConvAgent_ElevenLabs, Warning,
TEXT("[AUDIO-DBG] Role=%d splice discontinuity: |delta|=%d (prev=%d first=%d) count=%d — pop likely here."),
static_cast<int32>(GetOwnerRole()), Delta,
static_cast<int32>(LastEnqueuedSample), static_cast<int32>(Samples[0]),
AudioSpliceDiscontinuityCount);
}
else if (DbgVal >= 2)
{
UE_LOG(LogPS_AI_ConvAgent_ElevenLabs, Log,
TEXT("[AUDIO-DBG] Role=%d splice ok: |delta|=%d"),
static_cast<int32>(GetOwnerRole()), Delta);
}
}
LastEnqueuedSample = Samples[NumSamples - 1];
bHasLastEnqueuedSample = true;
}
}
{
FScopeLock Lock(&AudioQueueLock);
AudioQueue.Append(PCMData);
@@ -1908,6 +1944,7 @@ void UPS_AI_ConvAgent_ElevenLabsComponent::StopAgentAudio()
bPreBuffering = false; // Clear pre-buffer state on stop.
ClientReassemblyPCM.Reset(); // Client: drop any half-received message so it can't leak into the next turn.
AgentAudioEncodeLeftover.Reset(); // Server: drop the carried sub-frame tail so it can't glue onto the next turn.
bHasLastEnqueuedSample = false; // Debug: don't count the inter-turn splice as a discontinuity.
{
FScopeLock Lock(&AudioQueueLock);
AudioQueue.Empty();

8
Unreal/PS_AI_Agent/Plugins/PS_AI_ConvAgent/Source/PS_AI_ConvAgent/Public/PS_AI_ConvAgent_ElevenLabsComponent.h
View File

@@ -819,6 +819,14 @@ private:
// per turn (new-turn guard in HandleAudioReceived + StopAgentAudio()).
TArray<uint8> AgentAudioEncodeLeftover;
// Debug: detects waveform discontinuities at enqueue/message splices (the
// signature of the audio "pops"). Tracks the last enqueued sample; a large
// jump to the next block's first sample is flagged and counted. Toggle with
// ps.ai.ConvAgent.Debug.ElevenLabs. Reset in StopAgentAudio().
int16 LastEnqueuedSample = 0;
bool bHasLastEnqueuedSample = false;
int32 AudioSpliceDiscontinuityCount = 0;
// Pre-buffer state: delay playback start to absorb TTS inter-chunk gaps.
bool bPreBuffering = false;
double PreBufferStartTime = 0.0;