Transcription

POST /v1/audio/transcriptions turns speech into text: meetings, calls, voice notes, podcasts. Send an audio file or a URL and get back the transcript with word-level timings. The current model is grok-stt-1-0, billed per hour of audio.

Upload a file

curl https://api.cloakray.com/v1/audio/transcriptions \
  -H "Authorization: Bearer $CLOAKRAY_API_KEY" \
  -F "model=grok-stt-1-0" \
  -F "file=@meeting.wav"

import fs from "node:fs";
import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://api.cloakray.com/v1", apiKey: process.env.CLOAKRAY_API_KEY });

const transcript = await client.audio.transcriptions.create({
  model: "grok-stt-1-0",
  file: fs.createReadStream("meeting.wav"),
});

console.log(transcript.text);

Common formats work: WAV, MP3, OGG, Opus, FLAC, AAC, MP4 and M4A. Request bodies are limited to 64 MiB; for longer recordings, pass a URL.

Pass a URL

curl https://api.cloakray.com/v1/audio/transcriptions \
  -H "Authorization: Bearer $CLOAKRAY_API_KEY" \
  -F "model=grok-stt-1-0" \
  -F "url=https://example.com/podcast-episode.mp3"

Options

All fields are multipart form fields, so booleans are the strings true and false.

Field	What it does
`model`	Required.
`file` or `url`	Required, one of the two.
`language`	A language code such as `en`. Numbers and currencies are then written out properly.
`diarize`	`true` to label each word with its speaker.
`filler_words`	`true` to keep "uh" and "um".
`keyterm`	A name or term to recognise reliably, such as a product name. Repeat the field for several.

The response

{
  "text": "The balance is $167,983.15.",
  "language": "English",
  "duration": 3.45,
  "words": [
    { "text": "The", "start": 0.24, "end": 0.48 },
    { "text": "balance", "start": 0.48, "end": 0.96 }
  ]
}

start and end are seconds from the beginning. With diarize=true, each word also has a speaker.

The response is always this JSON. Plain text, SRT and VTT output via response_format are not offered; build subtitles from words if you need them.