Recorded Audio Processing Speed
See how long recorded audio took to process in production, from short clips to recordings lasting several hours.
Historical Processing Times
Up to 1 Minute of Audio
- Average, Seconds
- 10.01
- Completed Jobs
- 33,843
Over 1 to 5 Minutes of Audio
- Average, Seconds
- 12.22
- Completed Jobs
- 17,494
Processing Time by Audio Length
Historical averages for JSON transcription, with add-ons and streaming reported off.
| Audio Length | Completed Jobs | Average | Delay | Execution |
|---|---|---|---|---|
| Up to 1 | 33,843 | 10.01 | 5.98 | 3.70 |
| Over 1 to 5 | 17,494 | 12.22 | 6.06 | 5.83 |
| Over 5 to 15 | 6,828 | 18.14 | 6.26 | 11.53 |
| Over 15 to 30 | 3,299 | 26.28 | 6.66 | 19.23 |
| Over 30 to 60 | 1,316 | 36.71 | 7.57 | 28.73 |
| Over 60 to 120 | 108 | 52.14 | 7.63 | 44.11 |
| Over 120 to 240 | 3 | 85.31 | 5.19 | 79.75 |
Audio length in minutes. Processing times in seconds. Average covers internal processing time. Delay includes queueing and startup. Delay and execution are provider-reported averages. No matching jobs were recorded above 240 minutes.
Recent Median and Tail Timings
A separate six-day snapshot across all observed settings.
| Audio Length | Completed Jobs | Median | p95 |
|---|---|---|---|
| Up to 1 | 2,560 | 5.64 | 29.87 |
| Over 1 to 5 | 1,346 | 8.12 | 39.54 |
| Over 5 to 15 | 659 | 13.50 | 45.70 |
| Over 15 to 30 | 402 | 20.90 | 62.96 |
| Over 30 to 60 | 326 | 31.79 | 87.60 |
| Over 60 to 120 | 143 | 62.00 | 176.69 |
| Over 120 to 240 | 13 | 111.18 | 591.70 |
| Over 240 to 480 | 1 | 153.84 | 153.84 |
Audio length in minutes. Processing times in seconds. Median is p50. p95 covers 95% of measured completions. -- means no valid observations. These results use a different period and settings mix from the historical averages.
Recorded Job Outcomes
The same release group recorded 5,450 completed jobs and 37 failed jobs.
- Completed Jobs
- 5,450
- Failed Jobs
- 37
Frequently Asked Questions
Do these times include fetching audio and receiving the transcript?
The clock covers internal processing, from job-record creation after input fetching to completion handling before transcript storage. Fetching audio, webhook delivery, and retrieving the transcript add time outside these measurements.
Will my recordings take the same amount of time?
Processing time varies with audio length, content, requested features, and queue conditions. These results describe the production jobs in the stated window. They are not an SLA or a guarantee for an individual recording.
Which transcription settings do these benchmarks cover?
The primary historical averages include jobs with sound labels, speaker labels, word timestamps, and streaming reported off in JSON output. The recent table reports median and p95 figures across all observed settings. These production snapshots use different recordings and settings, so they cannot isolate the time an add-on adds.
Evaluate Speech is Cheap on Your Audio
Choose an API plan for your volume and evaluate processing time on your own recordings.
Methodology and Sources
Server-Side Elapsed Time
The clock starts when the service creates a job record, after input fetching and dispatch. It stops in the completion handler, before transcript storage, webhook delivery, or retrieval by your application.
Historical Release Group
The 2026-07-21 group includes equivalent releases 2026-07-15, 2026-07-16, 2026-07-21. The table copies published Bench averages for JSON output with sound labels, speaker labels, word timestamps, and streaming reported off. Bench treats absent flags as off and absent output format as JSON. Other settings and spoken language are unverified.
Observed jobs span through . These dates describe retained production traffic. They were not a predefined test window. The three jobs above 120 minutes provide limited evidence for longer recordings.
Historical averages are arithmetic means. Delay and execution averages omit missing values, but their valid sample counts are unavailable. This source has no historical percentiles or failed-job counts.
Recent Percentile Snapshot
This snapshot uses existing production records from six complete UTC days. The observation window was chosen before inspecting the timings. Recordings and requested settings vary between jobs, so the results describe this workload rather than a controlled test or an SLA.
The median is the 50th percentile. p95 is the time within which 95% of measured jobs finished. Percentiles use the nearest-rank method. Small groups do not establish stable tail percentiles.
All completed jobs in the measured release group have valid elapsed, queue-delay, and execution records. A repeated completion callback can overwrite the retained timing. The records do not distinguish the first attempt from later retries.
The full window contains 5,490 stored jobs. Two completions used earlier releases, and one failed job has no release identifier. These three jobs are outside the recent timing tables. The 37 failed jobs have no recorded audio length, so failure counts cannot be split by duration. Requests rejected before job creation are outside this snapshot. No stored jobs were canceled, pending, unknown, missing analytics, or status-mismatched.
Bucket lower bounds are exclusive and upper bounds are inclusive. The 4.5 to 5.5 minute and 55 to 65 minute configuration bands include both endpoints.
Configuration Details
Base defaults define a requested segment duration of 30 seconds, a minimum confidence of 0.5, and JSON output format. Base settings disable speaker labels, sound labels, word timestamps, streaming, and channel splitting.
Nearly all requests in this measurement window used automatic or unspecified language detection. One job requested English in the 55–65 minute cohort, and the 5-minute English-requested cohort had zero requests. Because requested language serves as a routing hint rather than verified spoken language, these measurements do not represent an English-only cohort.
Small sample counts provide descriptive historical snapshots rather than stable statistical guarantees, and small cohorts do not establish stable tail percentiles. For example, the base 55–65 minute cohort contains eight jobs and cannot characterize all hour-long audio. Similarly, the strict diarization-only default cohort recorded zero requests during this window, which reflects customer configuration choice rather than a lack of platform support.
| Configuration | Completed Jobs | Median | p95 |
|---|---|---|---|
| 4.5 to 5.5 Minutes All Observed Settings · No Language Filter | 145 | 9.74 | 41.81 |
| 4.5 to 5.5 Minutes All Observed Settings · English Requested No completed jobs were recorded in this duration bucket during the observation window. | 0 | -- | -- |
| 4.5 to 5.5 Minutes Base Defaults · No Language Filter | 133 | 9.74 | 41.81 |
| 4.5 to 5.5 Minutes Base Defaults · English Requested No completed jobs were recorded in this duration bucket during the observation window. | 0 | -- | -- |
| 4.5 to 5.5 Minutes Diarization Only · No Language Filter No completed jobs were recorded in this duration bucket during the observation window. | 0 | -- | -- |
| 4.5 to 5.5 Minutes Diarization Only · English Requested No completed jobs were recorded in this duration bucket during the observation window. | 0 | -- | -- |
| 55 to 65 Minutes All Observed Settings · No Language Filter | 62 | 48.18 | 104.94 |
| 55 to 65 Minutes All Observed Settings · English Requested | 1 | 30.63 | 30.63 |
| 55 to 65 Minutes Base Defaults · No Language Filter | 8 | 32.48 | 70.08 |
| 55 to 65 Minutes Base Defaults · English Requested No completed jobs were recorded in this duration bucket during the observation window. | 0 | -- | -- |
| 55 to 65 Minutes Diarization Only · No Language Filter No completed jobs were recorded in this duration bucket during the observation window. | 0 | -- | -- |
| 55 to 65 Minutes Diarization Only · English Requested No completed jobs were recorded in this duration bucket during the observation window. | 0 | -- | -- |
| All Recorded Durations All Observed Settings · No Language Filter | 5,450 | 9.29 | 66.55 |
| All Recorded Durations All Observed Settings · English Requested | 1 | 30.63 | 30.63 |
| All Recorded Durations Base Defaults · No Language Filter | 4,885 | 8.30 | 41.32 |
| All Recorded Durations Base Defaults · English Requested No completed jobs were recorded in this duration bucket during the observation window. | 0 | -- | -- |
| All Recorded Durations Diarization Only · No Language Filter No completed jobs were recorded in this duration bucket during the observation window. | 0 | -- | -- |
| All Recorded Durations Diarization Only · English Requested No completed jobs were recorded in this duration bucket during the observation window. | 0 | -- | -- |
Recent Delay and Execution Timings
The processing provider reports delay before execution, including queueing and startup. This does not cover every wait in the API request. Delay and execution percentiles cannot be added together. The processing provider reports execution time separately. It covers one part of the recorded job time.
Audio length in minutes. Delay and execution times in seconds.
| Audio Length | Queue and Startup Delay | Execution Time | ||||
|---|---|---|---|---|---|---|
| Jobs | Median | p95 | Jobs | Median | p95 | |
| Up to 1 | 2,560 | 2.37 | 25.88 | 2,560 | 2.39 | 6.88 |
| Over 1 to 5 | 1,346 | 2.51 | 33.11 | 1,346 | 4.67 | 10.41 |
| Over 5 to 15 | 659 | 2.59 | 26.66 | 659 | 9.06 | 21.46 |
| Over 15 to 30 | 402 | 2.83 | 22.42 | 402 | 16.00 | 36.36 |
| Over 30 to 60 | 326 | 1.01 | 16.40 | 326 | 26.88 | 66.91 |
| Over 60 to 120 | 143 | 0.86 | 10.70 | 143 | 57.88 | 158.70 |
| Over 120 to 240 | 13 | 0.86 | 11.01 | 13 | 106.27 | 589.62 |
| Over 240 to 480 | 1 | 0.97 | 0.97 | 1 | 151.32 | 151.32 |
Audio length in minutes. Delay and execution times in seconds.
| Configuration | Queue and Startup Delay | Execution Time | ||||
|---|---|---|---|---|---|---|
| Jobs | Median | p95 | Jobs | Median | p95 | |
| 4.5 to 5.5 Minutes All Observed Settings · No Language Filter | 145 | 1.22 | 30.73 | 145 | 6.55 | 13.23 |
| 4.5 to 5.5 Minutes All Observed Settings · English Requested | 0 | -- | -- | 0 | -- | -- |
| 4.5 to 5.5 Minutes Base Defaults · No Language Filter | 133 | 1.56 | 30.73 | 133 | 6.45 | 13.01 |
| 4.5 to 5.5 Minutes Base Defaults · English Requested | 0 | -- | -- | 0 | -- | -- |
| 4.5 to 5.5 Minutes Diarization Only · No Language Filter | 0 | -- | -- | 0 | -- | -- |
| 4.5 to 5.5 Minutes Diarization Only · English Requested | 0 | -- | -- | 0 | -- | -- |
| 55 to 65 Minutes All Observed Settings · No Language Filter | 62 | 0.85 | 12.33 | 62 | 44.83 | 104.16 |
| 55 to 65 Minutes All Observed Settings · English Requested | 1 | 14.61 | 14.61 | 1 | 15.79 | 15.79 |
| 55 to 65 Minutes Base Defaults · No Language Filter | 8 | 1.08 | 24.27 | 8 | 27.07 | 62.32 |
| 55 to 65 Minutes Base Defaults · English Requested | 0 | -- | -- | 0 | -- | -- |
| 55 to 65 Minutes Diarization Only · No Language Filter | 0 | -- | -- | 0 | -- | -- |
| 55 to 65 Minutes Diarization Only · English Requested | 0 | -- | -- | 0 | -- | -- |
| All Recorded Durations All Observed Settings · No Language Filter | 5,450 | 2.02 | 25.37 | 5,450 | 4.19 | 34.85 |
| All Recorded Durations All Observed Settings · English Requested | 1 | 14.61 | 14.61 | 1 | 15.79 | 15.79 |
| All Recorded Durations Base Defaults · No Language Filter | 4,885 | 2.69 | 28.09 | 4,885 | 3.72 | 18.44 |
| All Recorded Durations Base Defaults · English Requested | 0 | -- | -- | 0 | -- | -- |
| All Recorded Durations Diarization Only · No Language Filter | 0 | -- | -- | 0 | -- | -- |
| All Recorded Durations Diarization Only · English Requested | 0 | -- | -- | 0 | -- | -- |
The downloadable JSON contains the full aggregate results. The live Bench dashboard updates separately. Customer audio, transcripts, and account identifiers remain private.
Source JSON and Checksums
The frozen JSON files contain the published aggregates.
- Historical Bench JSONSHA-256 5f3533044787670b617e7fc49c6bf16f9bdacb2fe7876e75518f85a645365cc3
- Recent Percentile JSONSHA-256 3dda91dd5b9c306a40a626d29a19d4fc295b491f4f9b7106dedf89640888419c