Problem
At ~1:19 in sp_it_a_0001_00, the man is supposed to cut the woman off mid-sentence (BARGE_IN mix mode). Instead, the woman audibly stops speaking BEFORE the man starts, with a noticeable time gap. This sounds like two people politely taking turns, not an interruption.
Source: Listening test, 2026-05-03.
Expected behavior
In a barge-in:
- Woman is speaking
- Man's turn audio begins while woman is still audible
- Woman's audio is truncated shortly after the overlap starts
Likely cause
The SceneMixer BARGE_IN logic may be:
- Truncating the woman's turn at the planned onset of the man's turn, rather than letting them overlap
- Adding inter-turn silence/gap even in BARGE_IN mode
- Or the script generator may not be producing BARGE_IN mix mode annotations for this turn pair
Investigation needed
- Check the
MixedScene turn onset/offset values for the relevant turns — is there overlap in the timestamps?
- Check if the turn's
mix_mode is actually BARGE_IN or defaulting to SEQUENTIAL
- Review
SceneMixer.mix_barge_in() for how it handles the truncation point
Impact
P2. Affects realism of escalation scenes but doesn't block other work.
Problem
At ~1:19 in sp_it_a_0001_00, the man is supposed to cut the woman off mid-sentence (BARGE_IN mix mode). Instead, the woman audibly stops speaking BEFORE the man starts, with a noticeable time gap. This sounds like two people politely taking turns, not an interruption.
Source: Listening test, 2026-05-03.
Expected behavior
In a barge-in:
Likely cause
The
SceneMixerBARGE_IN logic may be:Investigation needed
MixedSceneturn onset/offset values for the relevant turns — is there overlap in the timestamps?mix_modeis actuallyBARGE_INor defaulting toSEQUENTIALSceneMixer.mix_barge_in()for how it handles the truncation pointImpact
P2. Affects realism of escalation scenes but doesn't block other work.