Google Speech-to-Text's 10 MB local limit and a 16 kHz mono Sume WAV
Google caps local audio at 10 MB for sync and async; Cloud Storage has no cap. A 16 kHz mono wav from Sume audio detach holds about five minutes in 10 MB.

Google Cloud Speech-to-Text accepts about one minute of audio on a synchronous request and up to 10 MB for local files. Asynchronous requests run to about 480 minutes, still with 10 MB for local files, with no limit for Cloud Storage URIs. A 16 kHz mono wav from Sume audio detach is about 1.92 MB a minute by arithmetic, so 10 MB holds roughly five minutes.
Google's figures are from its Speech-to-Text quotas page, read 2026-10-03. Sume's format options are from audio detach.
What the quotas say
The page splits limits by request type.
| Request type | Audio length | Size |
|---|---|---|
| Synchronous | About 1 minute | 10 MB for local files |
| Asynchronous | About 480 minutes | 10 MB local, no limit for Cloud Storage URIs |
| Streaming | About 5 minutes | 10 MB for local files |
Arithmetic for a Sume wav
Sume's default detach output is pcm_s16le: 16 bits is 2 bytes a sample. At sample_rate 16000 and mono, that is 32000 bytes a second, so 60 seconds is 1.92 MB and 300 seconds is 9.6 MB. This is general PCM arithmetic, not a figure from either vendor, and WAV headers add a few bytes.
Sume's maximum detach output is 900 seconds, which would be about 28.8 MB, so a full-length detach is three times too large for a local Google request. Take a range of 300 seconds or less, or send the file by Cloud Storage.
Detaching a minute for the sync call
Use range: { start, end } with a gap of 60 seconds or less. Keep channels at mono and sample_rate at 16000. Each job costs $0.01 and returns an audio_url on media.sume.com plus duration_seconds and source_duration_seconds.
Detach works only on media.sume.com sources, so the video must already be an artifact or asset in your workspace.
startpast the source end failsdetach_start_past_source.- A range longer than 900 s fails
audio_detach_range_empty. - A video with no audio fails
detach_source_has_no_audio.
When to use Cloud Storage instead
For a recording of an hour or more, the 10 MB local limit is the wrong path. Google's page says Cloud Storage URIs have no size limit on asynchronous requests, so upload the audio there and reference it. For Sume, anything over 30 minutes of source video is beyond the 1800 second source cap anyway.
Checking the numbers yourself
Download the detach result and read its byte size, then compare with Google's limit. The docs give a header size for neither side, and Google's page does not say whether MB means decimal or binary megabytes, so leave a margin of a few percent.
If you detach in 300 second ranges, each file is near the limit, so use 240 seconds for a safer margin.
A worked example
Take a 20 minute video. Detach it in ranges of 240 seconds, giving five files of about 7.7 MB each by the arithmetic above, each under the 10 MB local limit. Five jobs cost $0.05 at Sume's $0.01 per job.
Or upload the whole audio to Cloud Storage and use one asynchronous request, which Google says has no size limit for Storage URIs. That route needs a Google Cloud account and a bucket, but avoids the splitting.
Sources
Related posts
More in Media tools
- H3 Max 1080p image-to-video: check the first frames for a crop jump
Some 1080p MiniMax H3 Max image-to-video clips carry a baked crop jump after two frames. How to confirm it with video-frames and trim 2/24 s with source_in.
- Halloween teaser cut: Video trim start and duration, exact or keyframe
Cut a short Halloween teaser from a longer clip with Sume Video trim: start plus duration, frame-exact or keyframe copy, at a flat public rate per job.
- Holiday sale instrumental on Music Router: length in the prompt
Generate an instrumental sale-reel track with Sume Music Router: duration fields are rejected, so state length and structure in the prompt.
- How many audio tracks can one Instagram Reel hold? Up to 20
Meta says you can add up to 20 tracks to a single Reel. How to plan a multi-track sound bed ahead of time and build it in one Sume Timeline render.
Written by Sume