Repository navigation
AssemblyAI Dictation as a cloud transcription option #1697
Description
What are you trying to do?
Get text that's ready to send from a cloud model in one step. Today the cloud option (ElevenLabs) returns the words as spoken, filler and self-corrections included.
What would help?
AssemblyAI's Dictation API as a cloud option:
- One request:
POST https://dictation.assemblyai.com/v1/transcribewith the recording as a multipartaudiopart (WAV, raw PCM, MP3, M4A, Ogg/Opus, FLAC, WebM; 80 ms to 20 minutes) and anAuthorization: <api key>header. Raw PCM needssample_rateandchannelsin the config. - The response has
text(verbatim) andllm_response(filler words removed, punctuation, capitalization and self-corrections fixed). Usellm_response, falling back totextwhen it's empty. - An optional JSON
configpart takeskeyterms_prompt(names and jargon to spell right) andlanguage_codes. It covers 32 languages. - It isn't OpenAI-compatible, so it needs its own small request rather than the OpenAI-compatible path.
Disclosure: I work at AssemblyAI. If you have questions or want a hand with anything on our side, comment here or reach me on X at @dan_aai.
Activity
thdxr commentedon Nov 11, 2026
hex doesn't have a cloud transcription option right now. everything is transcribed on your device, and once the speech model is downloaded your audio never leaves the machine.
for cleaned-up, ready-to-send text today, a mode with rewriting turned on does that: it transcribes locally, then rewrites the result to drop filler words and fix self-corrections before it's pasted.
sending audio to a cloud service at all is a bigger call we haven't made yet, so it's the same decision as #765 and #603.
— from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊