Speech to text

POST /v1/audio/transcriptions takes an audio file and returns its text. The model is nur-v1, and it labels who spoke as part of every transcript.

curl https://api.eesi.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $EESI_API_KEY" \
  -F file=@interview.mp3 \
  -F model=nur-v1

import os
from openai import OpenAI

client = OpenAI(api_key=os.environ["EESI_API_KEY"], base_url="https://api.eesi.ai/v1")

with open("interview.mp3", "rb") as audio:
    result = client.audio.transcriptions.create(model="nur-v1", file=audio)
print(result.text)

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

const client = new OpenAI({ apiKey: process.env.EESI_API_KEY, baseURL: "https://api.eesi.ai/v1" });

const result = await client.audio.transcriptions.create({
  model: "nur-v1",
  file: fs.createReadStream("interview.mp3"),
});
console.log(result.text);

For a recording you want to search, cut or analyze later, use Audio intelligence instead. It keeps the recording and adds delivery, reactions and stems.

Parameters

Send the fields as multipart/form-data.

Field	Required	Notes
`file`	Yes	WAV, MP3, FLAC, OGG, M4A, WebM, or a video file with an audio track. Up to 1 GB.
`model`	Yes	`nur-v1`.
`language`	No	The spoken language, such as `en` or `ar`. Detected automatically when you leave it out.
`response_format`	No	`json` (default), `text` or `verbose_json`.
`timestamp_granularities[]`	No	`word`, `segment`, or both, with `verbose_json`.
`stream`	No	`true` streams results as they are decoded.

Pass language when you know it. Detection uses the first few seconds, so a recording that opens with music or a greeting in another language can be transcribed in the wrong one.

Speakers and timings

verbose_json adds segments with speakers, start and end times, and, if you ask for them, word timings:

{
  "language": "en",
  "duration": 3.2,
  "text": "The rain had not stopped for three days.",
  "segments": [
    {
      "id": 0,
      "start": 0.07,
      "end": 3.2,
      "speaker": "S01",
      "text": "The rain had not stopped for three days."
    }
  ]
}

To make captions, build SRT or WebVTT from segments.

Streaming long files

A long recording takes a while to decode. With stream=true the response is server-sent events: transcript.segment for each finished segment, and transcript.text.done with the full verbose_json result at the end.

import json, os

import httpx

with httpx.stream(
    "POST",
    "https://api.eesi.ai/v1/audio/transcriptions",
    headers={"Authorization": f"Bearer {os.environ['EESI_API_KEY']}"},
    files={"file": open("interview.mp3", "rb")},
    data={"model": "nur-v1", "stream": "true"},
    timeout=None,
) as response:
    for line in response.iter_lines():
        if line.startswith("data:"):
            event = json.loads(line[5:])
            if event["type"] == "transcript.segment":
                segment = event["segment"]
                print(f"[{segment['start']:.1f}s] {segment['text']}")

Each segment’s end divided by the file’s duration gives you a progress bar.

Limits


File size	1 GB
Length	About 90 minutes per file. Longer files are cut, and `verbose_json` returns `truncated: true`; split at a silence and send the parts.
Rate	60 requests per minute per organization

Usage is billed per minute of audio.

Get started

Quickstarts

Nur

Speech

Audio intelligence

Speech to text

Parameters

Speakers and timings

Streaming long files

Limits

​Parameters

​Speakers and timings

​Streaming long files

​Limits

Parameters

Speakers and timings

Streaming long files

Limits