Undertone · speech to text
Hears what’s under the words.
A transcript that keeps the feeling. Emotion, sounds and delivery land in the text where they happened, live as people speak.
joy
Honestly? I did not think it would work.
laughing
And then it just… did.
calm
So yes. Let’s ship it.
LIVE
HUMAN OR AI
FINAL IN ~264 ms
Try it
Say something. See what it heard.
Record or upload up to 15 seconds. You get the transcript with its feeling and sound tags.
Your audio goes to our API to be checked and is not stored by this demo.
What it hears
Three layers in every transcript.
The words
As people speak, with word timings. The final transcript lands about 264 ms after the speaker stops (median, October 2026). English today.
The feeling
Emotion and delivery per segment, so an agent can tell a “fine” that means fine from one that doesn’t.
Everything else
Laughs, sighs and breaths in place, a voice profile, and a human-or-AI score.
API
The request you already send.
Send the audio as the request body. Ask for verbose_json to get tagged segments, word timings and p_human.
curl "https://tts-api.yoob.com/v1/audio/transcriptions?response_format=verbose_json" \
-H "Authorization: Bearer $YOOB_API_KEY" \
-H "Content-Type: audio/wav" \
--data-binary @clip.wav