Multi-engine routing
Select a compatible model from a reviewed capability snapshot, with callback processing, polling fallback, and circuit-aware failover.
Transcription
A transcript you can trace back to the sound.
Route recordings through compatible speech engines, normalize word timing and speakers, and review the result in a playback-aware editor.
Pipeline contract
- Routing
- Capability-checked
- Timeline
- Word ranges
- Fallback
- Feature-compatible
Word-level timing
Keep normalized word ranges connected to playback so a reader can seek, inspect, and edit in context.
Speaker separation
Represent speaker turns as editable structure. Color supports the scan, while names and labels carry the meaning.
Inside the pipeline
Each transition stays explicit.
- 01
Prepare and route
Validate the media, create playback artifacts, and select an explicit hint or reviewed automatic-language route.
- 02
Transcribe and normalize
Accept signed callbacks or poll when needed, then normalize provider words without inventing missing text.
- 03
Review a stable draft
Publish one monotonic transcript revision and keep later retranscriptions separate until they are complete.
Coverage is a checked capability, not a headline number
Provider catalogs change. Echoryte routes only against a versioned capability snapshot that has passed the repository fixtures; a marketing language name is never treated as provider proof.
- Word timestamps and speaker labels are preserved only when the selected model supports the requested features.
- Automatic detection and an explicit language hint follow different routing paths.
- Failover uses a compatible provider or reports a stable error; it does not quietly drop diarization or change the requested tier.
- Names, uncommon terms, cross-talk, music, and poor microphones still need review against the audio.
Hear the source while you shape the text.
The anonymous trial physically clips media before transcription and returns only the first 90 seconds to this browser.
Try a recording