Speech inference / Iowa, United States · us-central1

Qwen3-ASR API

Multilingual audio transcription on a dedicated NVIDIA L4 in Iowa, United States · us-central1. Model: Qwen3-ASR-1.7B.

Authenticated private preview. Contact nathan@oruk.ai for regional access. Third-party marketplace availability and access keys depend on each platform’s approved integration.

Transcribe

curl https://qwen3-asr.oruk.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $ORUK_ASR_API_KEY" \
  -H "Idempotency-Key: my-recording-001" \
  -F "model=Qwen/Qwen3-ASR-1.7B" \
  -F "file=@audio.wav"

{"text":"The recognized speech appears here."}

Each upload supports up to 60 seconds and 25 MiB. JSON and plain text responses are available. The optional language field accepts an ISO language code; prompt accepts up to 1,000 characters. Realtime streaming and timestamps are not enabled.

Choose a region

Use the origin authorized for your key. Regional keys and request ledgers are independent. Keep transcription retries, idempotency keys and /v1/billing/costs queries on the same origin. A repeated idempotency key returns HTTP 409 because transcripts are not retained for replay. Requests are not automatically retried in another region.

Privacy and integration

Oruk’s gateway processes audio and transcription payloads in memory without persisting them. Request metadata is retained for billing and reliability. Third-party platforms apply their own upload, chat, routing and retention policies. Regional endpoints identify the inference compute location; they do not by themselves establish a contractual data residency guarantee.

OpenAI-compatible multipart transcription clients can use https://qwen3-asr.oruk.ai/v1. Preserve the unique Inference-Id response header for request and billing support.

OpenAPI specification · Region catalog · Privacy · Terms

Qwen3-ASR-1.7B is a Qwen model released under Apache 2.0. This service is operated by Oruk.