Realtime TTS-2 is live. Built for realtime conversation that feels human. Read the Realtime TTS-2 announcement

STT

STT Playground

Record live or upload audio to transcribe with Inworld STT

Try Inworld STT in the STT Playground before writing code. Test language hints and custom vocabulary, and watch Voice Profile analyze the speaker in real time.

STT Playground with model, language, custom vocabulary and speaker controls

Get Started

Go to Inworld Portal

In Portal, select Speech-to-Text from the left-hand side panel.

Choose a model

Open Model in the right-hand panel to select inworld/inworld-stt-1. See Supported models for its endpoints and language coverage.

Record or upload audio

Press the mic button to transcribe live from your microphone, or use the upload button to transcribe an audio file. The status bar shows Ready when the playground can accept audio and Listening… while recording.

Review the transcript

The transcript appears in real time as you speak. When you're done, use the copy button to grab the text, or Integrate to view the equivalent API request.

Advanced Features

To control transcription, try these settings:

  1. Language hint — The Language dropdown defaults to Auto-detect, which recognizes the spoken language automatically. Select a specific language to give the model a hint when you know the audio's language in advance — this generally improves accuracy, especially for short utterances. See Language Support for how hints behave.
  2. Custom vocabulary — To bias recognition toward names, jargon, or acronyms, enter them in Custom vocabulary and press Enter. Terms appear as chips, with the active term count below the input. This is a soft bias rather than a hard keyword lock — test it on the terms the baseline actually misses. Up to 100 terms, using letters, digits, spaces, and . , ' ( ) : ! ? ; - only — a term with any other character (such as &, _, or a curly quote) makes the request fail. The equivalent API field is prompts.
  3. Speaker diarization — Turn on the Speaker diarization toggle to label who said what in multi-speaker audio. The transcript prefixes each turn with a speaker label ([Speaker 1], [Speaker 2], …). The equivalent API fields are enableSpeakerDiarization together with includeWordTimestamps — speaker labels are returned on the word timestamps, so both must be enabled. See Speaker Diarization for how labels are assigned and its current experimental status on inworld/inworld-stt-1.
  4. Voice Profile — The panel on the right analyzes the speaker alongside transcription, showing Age, Accent, Voice Pitch, Vocal Style, and Emotion, each with a confidence score. It populates after transcription begins. See Voice Profiles for the full category reference.

From Playground to Code

Click Integrate at the top of the playground to generate an API request matching your current playground configuration: model, language hint, vocabulary, speaker diarization, and Voice Profile settings.

Next Steps

Ready to build? Here's where to go next.

Quickstart

Make your first STT API call in minutes.

Voice Profiles

Analyze speaker characteristics alongside transcription.

Language Support

Language coverage and hinting behavior.