How accurate is Zoom Scribe?

Zoom Scribe transcribes the 1,000 public Pipecat clips at 1.97% word error rate and the eight licensed Ocular recordings at 4.88%. Word error rate counts incorrect, missing, and extra words against the reference transcript, so lower is better.

How fast is Zoom Scribe in a live conversation?

First text arrives after 1.67s at the median, and the transcript is final 118ms after the speaker stops. Both are measured on the same 206 streamed turns as every other model, and the page shows P90 and P95 as well, which is what a caller meets on a bad turn.

How was Zoom Scribe measured?

Every model runs the same two datasets under the same conditions: 1,000 public clips from Pipecat's STT benchmark dataset for accuracy at scale, and eight licensed conversation recordings from Ocular for real speech. Latency comes from 206 streamed turns, using each provider's own supported finalization contract. Zoom Scribe returned usable text for 1,000 of the public clips.

Bench

Zoom Scribe

Every modelZoom

Accuracy vs. speed

Lower and further left is better

Best quadrantBest picks

Final-text delay P50 (ms)

Pipecat Dataset WER (%)

Zoom Scribe is 4th of 19 on the Pipecat Dataset and 17th of 19 on the Ocular Dataset.

1.97% Pipecat WER (4th) · 4.88% Ocular WER (17th) · 1.67s First text (12th) · 118ms Final text (7th)Likely range, 95% confidence: Pipecat WER 1.73 to 2.22%, Ocular WER 4.41 to 5.29%Published price $0.25 per hour of streamed audio. Scribe API Live, paid from prepaid credits. Source, checked Oct 8, 2026.

Last updated Sep 16, 2026Methodology by Dileep Chagam

Zoom Scribe against every other model

Lower is better on every measure. Hover a column for that model's full numbers.

Glossary: WER, TTFT, TTFS, P50

WER · Word error rate: Share of words the model got wrong: replaced, missing or added. Lower is better.
Pipecat Dataset: 1,000 short public clips from Pipecat's STT benchmark dataset.
Ocular Dataset: 8 recordings from 4 real phone conversations, licensed from Ocular.
Likely range · 95% range: Where the model's true score probably falls, given how much audio was tested. Overlapping ranges are too close to call.
TTFT · Time to first text: How long after the audio starts streaming until the first words appear.
TTFS · Final-text delay: How long after the caller stops until the full transcript arrives. A voice agent waits for this before it replies.
P50 / P90 / P95 · Typical, slow and slowest turns: Half of turns are faster than P50. Only 1 in 10 is slower than P90, and 1 in 20 slower than P95.

Where Zoom Scribe struggled

On real calls, dropped words are its biggest source of error. On clean clips, wrong words lead.

Dropped word

What the speaker said, with Zoom Scribe's mistakes marked: I need to find a comprehensive guide on how to properly care for a bonsai tree indoors, specifically focusing on watering techniques, soil composition, and seasonal pruning for a ~~ficus~~ (left out) species.

Everything Zoom Scribe wrote

I need to find a comprehensive guide on how to properly care for a bonsai tree indoors. Specifically focusing on watering techniques. soil composition, and seasonal pruning for a species.

Pipecat Dataset clip pipecat-9ee9d60f-f794-36f2-cc7b-53bfb396ef1d

Wrong word

What the speaker said, with Zoom Scribe's mistakes marked: The instructions for assembling this ~~bookshelf~~ written as book shall are just pictures with no text and I've gotten to a step where I'm not sure which direction a certain panel is supposed to face and

Everything Zoom Scribe wrote

The instructions for assembling this book shall. are just pictures with no text, and I've gotten to the step where I'm not sure which direction a certain panel is supposed to face, and…

Pipecat Dataset clip pipecat-ecae6e0e-8bad-483f-1e0b-552bc91199c4

Marked: words Zoom Scribe left out or got wrong. Audio from the Pipecat Dataset, under CC BY 4.0, re-encoded as MP3.

TTFT and TTFS at P50, P90 and P95

Timed on the same 206 streamed Ocular turns as every other model. P90 and P95 show what a caller meets on a slow turn.

TTFS, final text

How long after the caller stops until the full transcript arrives. A voice agent waits for this before it replies.

20ms50ms100ms200ms500ms1s2s

Typical, P50: 118ms; 7th fastest of 19
Slow, P90: 216ms; 9th fastest of 19
Slowest, P95: 311ms; 10th fastest of 19

TTFT, first text

How long after the audio starts streaming until the first words appear.

200ms500ms1s2s5s10s20s

Typical, P50: 1.67s; 12th fastest of 17
Slow, P90: 9.02s; 16th fastest of 17
Slowest, P95: 14.86s; 16th fastest of 17

Every other model's typical turn, on the same axis.

1,000 Pipecat clips · 8 Ocular recordings · 206 turns timed · 0 failed requests

Zoom Scribe FAQ

Where does Zoom Scribe place in the benchmark?

Zoom Scribe places highest on the Pipecat Dataset, 4th of 19 at 1.97%. Its weakest placing is the Ocular Dataset, 17th of 19 at 4.88%.

Zoom Scribe places highest on the Pipecat Dataset, 4th of 19 at 1.97%. Its weakest placing is the Ocular Dataset, 17th of 19 at 4.88%.
Zoom Scribe transcribes the 1,000 public Pipecat clips at 1.97% word error rate and the eight licensed Ocular recordings at 4.88%. Word error rate counts incorrect, missing, and extra words against the reference transcript, so lower is better.
First text arrives after 1.67s at the median, and the transcript is final 118ms after the speaker stops. Both are measured on the same 206 streamed turns as every other model, and the page shows P90 and P95 as well, which is what a caller meets on a bad turn.
Every model runs the same two datasets under the same conditions: 1,000 public clips from Pipecat's STT benchmark dataset for accuracy at scale, and eight licensed conversation recordings from Ocular for real speech. Latency comes from 206 streamed turns, using each provider's own supported finalization contract. Zoom Scribe returned usable text for 1,000 of the public clips.