Speech to text models that handle noisy audio and identify speakers
The Hook
“Introducing ElevenLabs Scribe V3 and Scribe V3 Live. The most accurate Speech to Text models in the industry.”
Feature Teaser
“Scribe V3 brings a major leap in diarization, with up to 17% improvement in audio with three speakers, and 27% fewer word errors across 102 languages.”
Feature Teaser
“And you get more control over the transcript itself. Set formatting rules in plain English and choose how numbers come back: words or digits.”
Feature Teaser
“Scribe V3 Live for the first time supports diarization. Its turn detection now understands the context of the whole conversation, and it's more accurate than ever, with 43% fewer word errors, beating every other real-time model we tested.”
Product Reveal
“With attention to detail like never before. Thanks for calling Harbor Mutual, Dana speaking. How can I help? Hi, I'm calling to file a claim. I was rear-ended on Route 9.”
Product Reveal
“Sorry to hear that. Are you okay? Yes, just shaken up. Policy is HM-4470219.”
Product Reveal
“Built for global teams, Scribe V3 follows every language switch without missing a beat.”
Product Reveal
“In some scenarios, the details are crucial. Scribe V3 pays close attention and catches every note, word, and number precisely. Note for tomorrow: Level 4 slab, gridline C-7. Rebar spacing looks like 150 millimeters, not the 125 on the drawings. Check with the engineer before the 6:30 pour on Friday.”
Call to Action
“Try both models today. Scribe V3 and Scribe V3 Live are now available across ElevenLabs.”
Synthesizable in Remotion · key components of the clean-saas-benchmark-reveal template.
Displays transcription text word-by-word with subtle vertical slide and opacity fade-in.
Use useCurrentFrame to interpolate opacity from 0 to 1 and translateY from 10 to 0 using spring physics. Append words sequentially based on audio timestamps.
Horizontal metric bar dynamically growing from width zero to target percentage with leading value label.
Map progress using interpolate with extrapolateClamp from 0 to targetPercentage, driving CSS flex/width styling.
Expanding outline contours pulsing outward to represent speech processing.
Create layered SVG rounded rectangles, scaling stroke-width and scale with modulo frame math or continuous interpolate loops.
Purpose. Establish authority and introduce model names immediately.
Execution. Bold center typography over dynamic liquid blue gradient backdrop.
Purpose. Demonstrate quantitative superiority over prior versions and competitors.
Execution. High-contrast stat callouts (17%, 27%, 43%) paired with minimal bar charts.
Purpose. Showcase real-world enterprise edge cases including formatting, live dialog, and accents.
Execution. Interactive UI panels typing out phone calls, construction logs, and bilingual meetings.
Purpose. Prompt developer and enterprise adoption.
Execution. Return to hero gradient canvas with clear closing callout and logo lockup.