surfacetransformationtranscribe
Transformationscalar · returns varchar

TRANSCRIBE

Speech-to-text via Whisper large-v3-turbo

Per-row — runs once for each row.

transformationllmtext

Syntax

TRANSCRIBE({{ audio }})
TRANSCRIBE({{ audio }}, '{{ language }}')

Arguments

nametypedescription
audioVARCHAR
language(optional)VARCHAR

About

Speech-to-text — turns an audio column into a text column via the specialist zoo's Whisper large-v3-turbo. Once transcribed, every other text operator (MEANS, ABOUT, EXTRACTS, SENTIMENT, CLASSIFY, SUMMARIZE, MEANING, THEMES, …) composes naturally on top. Accepts any of: - https:// or http:// URL - Local path visible to the gateway - data:audio/...;base64,... URI - Raw base64-encoded audio bytes Handles arbitrary-length audio via the gateway's chunked inference path (30s windows, beam batching), so hour-long call recordings or meeting transcripts work out of the box. For LLM-style audio Q&A (e.g., "what did the caller ask about?" without the raw transcript), use an LLM cell on top of the TRANSCRIBE output. This cascade is purely the speech-to-text conversion.

Nearby rabbit holes

same domain
Climb back to The Looking Glass