Should I transcribe both speakers during an overlap?

Where the words are recoverable, yes, as separate labelled turns. Where they are not, mark the overlap rather than leaving a silent gap.

Does better audio fix overlap?

Separate microphones per speaker do, because each track carries one voice. A single room microphone does not, regardless of quality.

Why did the speaker count change after an overlap?

Diarization often mis-clusters across an overlap, merging two speakers or inventing a third. It is a normal correction in review.

Can separate microphones eliminate the problem?

They largely solve attribution, because each track carries one speaker. Recovering words that were spoken simultaneously still depends on how much bleed there is between the tracks.

Definition

What Is Overlapping Speech?

VideoText editorial team. Updated September 13, 2026.

Direct definition

Overlapping speech is two or more people talking at the same time. It is one of the hardest conditions for automatic speech recognition and for diarization.

The sections below explain how overlapping speech is used in transcription and subtitle work, what people often mix up, and which VideoText workflow applies when the next step is a file or review pass.

Key takeaways

Style guides often use [crosstalk] or parallel speaker lines.
Distant-microphone meeting sets show much higher WER than headset audio.
Do not pretend an ASR draft resolved overlap if the audio is unintelligible.

Why it is hard

Models trained mostly on single-speaker read speech under-represent overlap. The Whisper paper’s AMI distant-microphone WER (36.4% for Large V2) is a public reminder that meetings are not LibriSpeech.

Why recognisers struggle with it

Most speech recognition assumes one voice at a time. When two people speak together the acoustic signal is a mixture, and the model typically transcribes whichever voice is louder while dropping the other entirely, without indicating anything was lost.

Diarization degrades at the same moment, because the clustering that assigns audio to speakers has no clean single-speaker segment to work from. Overlaps are therefore where both the words and the attribution tend to fail together.

How style guides ask you to mark it

Conventions vary. Some guides use a dedicated tag for simultaneous speech, some ask for both turns to be transcribed in full and marked, and some accept transcribing the dominant speaker with a note that others were talking.

What is consistent across guides is that silently dropping the second speaker is an error. If you cannot make out the overlapped speech, the expectation is an inaudible tag rather than an omission.

Frequently asked questions

Should I transcribe both speakers during an overlap?: Where the words are recoverable, yes, as separate labelled turns. Where they are not, mark the overlap rather than leaving a silent gap.
Does better audio fix overlap?: Separate microphones per speaker do, because each track carries one voice. A single room microphone does not, regardless of quality.
Why did the speaker count change after an overlap?: Diarization often mis-clusters across an overlap, merging two speakers or inventing a third. It is a normal correction in review.
Can separate microphones eliminate the problem?: They largely solve attribution, because each track carries one speaker. Recovering words that were spoken simultaneously still depends on how much bleed there is between the tracks.

Related terms

Sources

OpenAI: Whisper paper, AMI rows in Table 2