Sends your audio to Subtitld Cloud, where AssemblyAI turns it into timed subtitles, each with the speaker who said it. It can't be used yet: Subtitld Cloud isn't open to the public.

Not usable yet. Everything goes through Subtitld Cloud, and Subtitld Cloud isn't publicly running yet: its address, cloud.subtitld.org, answers "not found", so every transcription stops with an error. Once it opens you will also need a Subtitld Cloud API key and credit. If you have access to a running Subtitld Cloud server, set SUBTITLD_CLOUD_BASE_URL to its address before you start Subtitld.

What it does

Who said it

With Detect speakers (diarization) on, each subtitle gets its speaker: A, B, C…

Sentences, not fragments

One subtitle per sentence, as AssemblyAI splits them, or one per speaker turn.

All at once

The progress bar moves while the job runs, but the subtitles arrive together when it ends.

Cancel releases the credit

Cancelling also cancels the job on Subtitld Cloud, so the credit held for it is released.

Your audio leaves your computer. The add-on turns the sound into a small Opus file (mono, 24 kbps, about 10 MB an hour) and uploads it to Subtitld Cloud with your key. Subtitld Cloud passes it on to AssemblyAI under its own AssemblyAI account; the job carries the model, language and options you chose. The add-on deletes its copy when the run ends. For material that must stay on your computer, use Whisper or Vosk.

Upgrading from 26.03? AssemblyAI was built in then and used your own AssemblyAI key. This add-on doesn't: it signs in to Subtitld Cloud with a Subtitld Cloud key, and never talks to AssemblyAI itself.

Options

In Settings › Add-ons › AssemblyAI (Subtitld Cloud) › Configure.

Subtitld Cloud API key

Default: empty

Your Subtitld Cloud key, not an AssemblyAI key. Leave it empty to use the key from the Subtitld Cloud tab.

Model

Default: subtitld-cloud:assemblyai/best

Best is the most accurate. Nano is faster and costs less.

Detect speakers (diarization)

Default: on

Tags each subtitle with the speaker who said it: A, B, C… Adds a little time on the server.

One cue per sentence

Default: on

On: one subtitle per sentence, as AssemblyAI splits them. Off: one per speaker turn, so longer subtitles.

You can also paste the key once in the Subtitld Cloud tab of the global settings; the add-on uses it when its own field is empty.

Languages

Regional tags go to AssemblyAI as their two-letter language: pt-br is sent as pt, en-gb as en.

en
en-us
en-gb
en-au
es
fr
de
it
pt
pt-br
nl
hi
ja
zh
fi
ko
pl
ru
tr
uk
vi

Platforms

System	Download	Tested
Linux x86-64	17.7 MB	In CI, against a stand-in cloud
macOS (Apple silicon)	14.5 MB	Built in CI, not run
Windows x86-64	8.6 MB	Built in CI, not run

Before each release, CI runs the tests on Linux, driving the add-on against a stand-in for Subtitld Cloud. The macOS and Windows downloads are built there but not run.

Troubleshooting

bad_paramsSomething to fix on your side; the message says what: no API key or a wrong one, an empty balance (with a link to top up), audio too large to accept (transcribe a shorter range), or no Subtitld Cloud at the address in use, which is the case until it opens.
model_missingSubtitld Cloud doesn't offer the chosen model. Choose the other one in Model.
network_unavailableSubtitld Cloud couldn't be reached, was busy, or had a server error. Try again in a moment. While a job is running, the add-on retries on its own first.
internalThe job failed on the server, the audio couldn't be encoded, or no result came within an hour. If the job had started, check your Subtitld Cloud dashboard before you run it again: it may have finished, and been charged.

License

Add-on

Apache-2.0

Service

AssemblyAI terms

Reached through Subtitld Cloud

The add-on's code is Apache-2.0. Using the transcription service is governed by AssemblyAI's own terms.

AssemblyAI (Subtitld Cloud)