Mixed measurement evidence

Set report

Thursday late show

8.1 minutes analyzed by automated response detection.

4.8detected responses / min

Check the tape before acting.Some detections rely only on speech pauses. Verify those moments against the tape before changing material.

Measurement receipt

What supports this report

39 detections ÷ 8.1 min = 4.8

Audio-backed20crowd-audio bursts

Transcript markers3may overlap audio events

Gap-only19verify these against the tape

Support coverage51%not an accuracy score

Analysis window8.1mof 9.0m recorded

Mixed measurement evidence. Some detections rely only on speech pauses. Verify those moments against the tape before changing material.

Detected responses39automated candidates, not human labels

Avg response2.7saverage detected-event duration

Longest response7.2slongest detected event in this set

Typical intensity58/100Mid · relative to this set only

Timing opportunity · check the tape

Possible talk-through moments

Three signals agreed: sustained audio + measured speech + laugh marker

3moments to review

You got a response. Now check whether you gave it room.

A sustained response overlapped your next words at these timestamps. That can mean you moved on early—but a tag, crowd-work exchange, close mic, or imperfect transcript timing can look the same. Listen before changing the bit.

5.9s estimated overlap8% of detected responsesReview not a fault score

012:513.1s response candidate72% speech overlapCheck your original recording
025:122.4s response candidate65% speech overlapCheck your original recording
037:102.0s response candidate55% speech overlapCheck your original recording

Why this is conservative: LaughSignal only surfaces a moment when audio, speech timing, and a machine laugh marker agree. It stores the timing measurements—not extra joke text—and never changes your LPM.

Unlocked this set

Achievements

+75Laugh Score

Hat TrickThree or more Bankers in a single set.

On the RiseYou beat your previous set's LPM.

Personal BestA new all-time LPM high.

See your whole trophy case

Every detection

Response timeline

Taller = longer detected event

Peak detections

Top 3 detected moments

Ranked within this recording by duration and relative audio intensity

3:36

7.2sdetected response

Near this set's peak

Audio-backed

“That is not a family group chat. That is a hostage negotiation with birthday GIFs.”

crowd-audio burstspeech pausetranscript laugh marker

8:17

6.1sdetected response

Near this set's peak

Audio-backed

“My smartwatch congratulated me for standing up. At my age, honestly, fair.”

crowd-audio burstspeech pausetranscript laugh marker

1:36

5.8sdetected response

Near this set's peak

Audio-backed

“I said, if the algorithm knows me that well, it should at least pay half the rent.”

crowd-audio burstspeech pausetranscript laugh marker

Moment by moment

Your set, reaction-tiered.

1:36Bankerstrongest response

…if the algorithm knows me that well, it should at least pay half the rent · 15-word setup

5.8s detected

2:13Workhorseshort response

…airport people clap when the plane lands like the pilot just parallel parked · 13-word setup

2.6s detected

2:38Hitterclear response

…I told the barista my name and she still spelled it like a ransom note · 15-word setup

3.7s detected

3:36Bankerstrongest response

…that is not a family group chat, that is a hostage negotiation with birthday GIFs · 15-word setup

7.2s detected

5:00Reviewlong response gap

…the bit about my gym membership ran 46 seconds without a detected response · 13-word setup

46s response gap

6:28Hitterclear response

…my landlord calls it 'cozy' — cozy is a word for small that pays no rent either · 17-word setup

4.9s detected

8:17Bankerstrongest response

…my smartwatch congratulated me for standing up — at my age, honestly, fair · 13-word setup

6.1s detected

Before this set’s strongest detected responses, the machine transcript averaged ~14 words. Verify the boundaries before treating that as a writing pattern.

Tiers are automated and set-relative. They classify detected response duration—not writing quality, audience intent, or whether a joke is proven across rooms.

Set #6 · your trajectory

How is this set trending?

This set4.8LPM

vs last set▲ 0.64.2 → 4.8

All-time best4.8LPM

Career Bankers14strongest-tier detections

Trend note: recording quality changes the detector. Compare sets recorded with a similar mic position, room, and crowd level.

🔒 Analyst

6 sets tracked — and counting.

Analyst charts every set over time, follows recurring bits night to night, and surfaces response patterns worth checking against the tape. Turn consistent measurements into a trajectory.

See your Analyst trajectory

Machine transcript · verify before quoting

Automatic transcription can mishear punchlines, names, audience noise, and speaker changes. It supports navigation; it is not a verbatim record.

…so I download the meditation app, right, and it opens with a notification: 'You haven't meditated in 14 days.' That is not mindfulness, that is my mother with a subscription model. And I said, if the algorithm knows me that well, it should at least pay half the rent. [laughter] My landlord calls the place cozy. Cozy is a word for small that pays no rent either. [laughter] I fly home for the holidays, the whole plane claps when we land, like the pilot just parallel parked a bus…

How to read this

Turn the numbers into your next set.

Detected responses per minute (LPM)

Automated response candidates divided by analyzed minutes. It is most useful as your own trend across similarly recorded sets—not as a universal comedy benchmark.

Reaction tiers → what to review

Banker, Hitter, and Workhorse rank response duration within the detector’s thresholds. Review flags a long response gap. None of them grade the writing itself.

Duration, intensity & setup

Duration is the measured event length. Intensity is normalized within this recording, so its loudest audio event is always 100. Setup counts come from a machine transcript and should be checked.

Evidence & limitations

Audio-backed events carry the strongest direct signal. Transcript markers are machine-generated. Gap-only events are clues, not confirmed laughs. Support coverage describes the inputs; it is not an accuracy percentage.

Possible talk-through → review timing

This optional note appears only when sustained audio, measured speech, and a laugh marker overlap. It points you to the tape; it does not decide whether a tag, riff, or transition was intentional.

Do next: verify the top three against the tape, review one long response gap, and record your next set the same way so the comparison is honest. Track the pattern over time in Analyst.

Straight with you

Measurement, with receipts.

LaughSignal combines crowd-audio bursts, speech pauses, and machine-transcript markers to find response candidates. It is automated measurement—not human judgment, proof of audience intent, or a grade on your writing. The evidence receipt shows exactly what supported the result.

Private processor · pipeline 1.4.0 · transcript model whisper.cpp-small.en

Analyze your set →