The audio analysis API for developers. POST a file or share a recording link. Define the fields you need, get transcript, sentiment, and more back as JSON in seconds.
HOW IT WORKS
List the fields you want back, like transcript, sentiment, speaker count, or topics. You choose them, the API fills them in.
POST a file directly, or generate a recording link. Users record or upload from any browser, no app needed.
A webhook fires the moment analysis finishes, with the transcript and every field you defined.
ONE SCHEMA ENGINE
Define the fields you need. The API returns exactly those, structured as JSON, on every recording.
Full text, timestamps, speaker turns.
Number, turn-taking, diarization labels.
Positive, negative, frustrated, satisfied.
Detection across 50+ languages.
What was discussed, in your own categories.
Promises, requests, follow-ups.
Silence, music, background noise, interruptions.
Define a field, get it back.
USE CASES
Every call analyzed the moment it ends. Extract sentiment, reason for call, resolution status.
Discovery calls turned into CRM updates without rep effort. Extract deal stage, objections, next steps.
Sample 100 percent of recordings instead of 2 percent. Extract script adherence, prohibited words.
Patient intake calls and symptom voicemails. Extract chief complaint, duration, urgency.
Give your voice agent structured context from every call it handles.
Recorded internally, summarized instantly. Extract attendees, decisions, action items.
PRICING
FAQ
Transcription gives you the words. We give you structured data. Define a schema, sentiment, action items, topics, anything, and the API returns the transcript plus those fields in the exact JSON shape your app expects.
MP3, WAV, M4A, AAC, OGG, and FLAC. Files up to 500 MB per request on Pay as you go, unlimited on Pro and Enterprise.
50+ languages, including English, Spanish, French, German, Mandarin, Japanese, and Arabic. Auto-detect, or pass a language hint for higher accuracy.
Yes. The API distinguishes speakers automatically, labels them, and attaches turn-taking metadata. Define speakers in your schema to get a count or per-speaker fields.
Most extractions finish in under 10 seconds. A 1-minute recording typically returns in 2 to 4 seconds, and the webhook fires the moment it is ready.
Word accuracy is typically 92 to 98 percent on clean audio. Every extraction includes a confidence score per field, so you can auto-accept high-confidence results.