Speech-to-Text Specialist

Transcribe audio to text with OpenAI Whisper.

Overview

• Endpoint(s): stt-public (Whisper small) · stt-turbo (Whisper large-v3-turbo)
• Model (stt-public): openai/whisper-small
• Capability: audio.transcribe
• Backend: specialist (GPU; falls back to CPU)
• Price: per request, USD — see the catalog for the current rate

When to use

✅ Meeting and call transcription, voice commands, subtitling, and audio indexing. stt-turbo (large-v3-turbo) is the higher-accuracy option; stt-public (small) is lighter and faster.

❌ Real-time streaming transcription — requests transcribe a complete audio clip, not a live stream.

Request / response

Pass an audio_url (the node downloads it and forwards the bytes) — or audio_base64 — plus an optional language hint. You get back the transcribed text and per-segment timings.

from colabhive import ColabHive

client = ColabHive(api_key="hive_...", account_id="YOUR_ACCOUNT_ID")

result = client.endpoints.infer(
    endpoint_id="stt-turbo",   # or "stt-public"; SDK resolves the name to a UUID
    input_data={"audio_url": "https://example.com/meeting.wav", "language": "en"},
)
data = result["result"] if result.get("status") != "queued" \
    else client.endpoints.get_task(result["task_id"])
# data → {"text": "...", "language": "en", "segments": [...], "inference_time_ms": 1830.0}

Field	Type	Description
audio_url	string	Audio URL — fetched and base64-encoded by the platform
audio_base64	string	Alternative: send base64 audio bytes directly
language	string	Language hint, e.g. en, es (optional; Whisper auto-detects)

You can also upload a local file with client.endpoints.upload_input(...) and pass the returned URL. The authoritative schema is GET /api/builder/v1/endpoints/{id}.

Transcription can be slow — prefer async

Longer clips take time, and a cold model pays a load cost. Poll GET /tasks/{task_id} if the call returns {"status": "queued"}. See Model Readiness & States.

Tips

• WAV/FLAC/MP3 all work; clean audio transcribes far more accurately than noisy recordings.
• Provide the language hint when you know it to skip auto-detection.

Next steps

• All Specialists · OCR
• Inference API

---

Authors: José Luis Minich, Maximiliano Lucius.