Introduction
utter is a speech to text API for Indian speech. The first model targets the way people actually talk at work: Hindi and English mixed inside a single sentence.
Invoice share kar diya hai, meeting 5 baje start hoti hai.
That sentence is one utterance, not two. Systems that expect a single language per
request tend to either romanise the Hindi badly or drop it. utter treats
hi-IN+en-IN as its own language setting.
What you can do
Section titled “What you can do”Stream. Open a WebSocket, push audio as you capture it, and read partial
results while the speaker is still talking, then final results as segments close.
See Streaming over WebSocket.
Transcribe a file. POST /v1/transcribe with an audio file and get one
response back. See Batch transcription.
Choose your latency. Every request names a tier. The tier sets how much
right context the encoder is allowed to wait for, which is both the latency knob
and the rate. See Latency tiers.
Pay from a prepaid balance. You buy credits up front, and every second of audio debits them as it is transcribed. There is no invoice. See Pricing and credits.
How the pieces fit
Section titled “How the pieces fit”Your credential is an API key that talks to the utter API over HTTPS and WSS. The API is the only thing that speaks to the serving layer that holds the models. No customer credential ever reaches a GPU, and the serving layer is not publicly addressable.
your app --(Authorization: Bearer utt_live_…, wss)--> api.utter.ccShape of the API
Section titled “Shape of the API”| Base URL | https://api.utter.cc |
| Streaming | wss://api.utter.cc/v1/stream |
| Auth | Authorization: Bearer utt_live_… |
| Audio in | float32 mono little endian PCM |
| Metering | per second of audio, no rounding up to the minute |
What is not here yet
Section titled “What is not here yet”There is no CLI, and no SDK published to PyPI or npm. Every example in these docs
uses ordinary HTTP and WebSocket clients (websockets and requests in Python,
the browser’s built in WebSocket), because those work today.
The generated REST reference is not published yet either. See REST reference for what is coming and where it will live.
- Quickstart gets you a key and a first transcript.
- Authentication covers live keys, test keys and rotation.
- Languages lists what is available and what is in training.