Orukeet · Preview

Get the words.
Keep the wait short.

English transcription for dictation and short recordings, at $0.027 per audio hour. Stream audio as someone speaks, then receive the finished transcript when they stop.

Try Orukeet Compare with Whisper

One key. Choose your tasks.

Use your Oruk key with the transcription endpoint. Omit all optional flags for the shortest response time, or enable the tasks you need. Speaker results include a transcript for each turn; speaker timestamps use seconds.

Orukeet with optional tasks

curl https://speech-api.oruk.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $ORUK_API_KEY" \
  -F model=oruk-orukeet \
  -F file=@audio.wav \
  -F emotion_detection=true \
  -F diarize=true

This example uses your Orukeet allowance at $0.01245 per audio minute. Usage beyond the allowance is billed at the same rate. Transcription alone is $0.00045 per minute. Emotion scores describe vocal delivery, not a person’s inner state.

Send audio while it’s being recorded

Preconnect to the WebSocket endpoint, wait for ready, send mono 16 kHz PCM16 frames, and send {"type":"commit"} at the end. Orukeet returns final text after the utterance; it does not emit incremental word hypotheses.

Configure optional tasks before sending audio. With tasks enabled, a transcript event arrives first and final includes the completed task results. Browser applications obtain a short-lived, single-use token from their authenticated backend.

Read the streaming protocol

Prefer to run it on your own hardware?

The downloadable Orukeet model transcribes 25 languages locally without an Oruk account. That local model is separate from this hosted English API and does not include its optional emotion or speaker-diarization tasks.

Run the local Python, OpenWhispr or ONNX example

The guide includes saved outputs, installation checks and an independently published Emacs integration. For the hosted API, use the measurements below and the reproduction protocol to inspect the request-level evidence.

Measured over the public API

Median final-audio-to-text latency: 97 ms
Median transcription processing, streaming run: 20 ms
Sustained direct REST requests, 8 concurrent: 6.6/s

September 12, 2026. Native Oruk key, one client, one warm A100 in the US. Streaming: 60 real-time replays, p95 119 ms after the final audio frame; connection setup and recording time excluded. Throughput uses a one-minute direct REST run. REST includes upload, authorization, credit reservation, processing, and response. Base transcription only.

See latency distributions and throughput

Pay for the tasks you need

Included in every plan. Transcription and selected tasks draw from your monthly Orukeet allowance at the rates below. Extra usage is billed monthly within your spending cap.

TranscriptionEnglish text, with file uploads or PCM streaming.: $0.00045 / audio min
Emotion detection15 emotion scores and labels from vocal delivery.: +$0.0080 / audio min
Speaker diarizationSpeaker labels and transcripts for each turn.: +$0.0040 / audio min

No audio-duration minimum. Each request’s total charge rounds up to one microdollar. Requested task failures return an error and do not charge the request. Optional tasks take additional processing time.

Choose a plan

Limits you can build around

Audio: English; at most 60 seconds per utterance
Upload: 4 MiB for the complete multipart request
Concurrency: 8 active requests or recordings per organization by default
Capacity: One US serving region; busy requests receive an explicit retry response

WAV, FLAC, MP3, M4A, OGG, and WebM are accepted. Mono 16 kHz PCM16 WAV or FLAC avoids format conversion. Latency varies with recording length, network distance, load, and selected tasks.

Try Resonance 2 early.

Explore oruk