Transcribe an iPhone voice memo (m4a) with an API: audio_url, limits

To transcribe an iPhone voice memo with Sume STT, put the .m4a at a public HTTPS URL and call POST /v1/stt-1.0/transcribe with it as audio_url. The OpenAPI examples use .m4a URLs, and the request validation checks the URL, not the file format. Each request covers at most 600 seconds (10 minutes) and is priced at $0.01 per audio minute.

Read on 2026-10-03: the Sume API reference for STT 1.0, the media-imports schema, the Media inputs docs page and the pricing catalog in the Sume repository.

What must audio_url be, and is m4a supported?

Verified in the request schema: a string of at most 2,048 characters, HTTPS, no username or password, no port and no private or local host. The field description says to prefer a Sume media or attachment URL. The published STT examples include https://media.sume.com/artifacts/example/clip.m4a, so m4a is a shape the API documents.

What I could not verify: a list of accepted codecs. The check stops at the URL, so a format problem would surface as a failed job. If an m4a fails, convert it first and send the converted file, for example ffmpeg -i memo.m4a -ac 1 -ar 16000 memo.wav.

Can media-imports host the memo?

No. POST /v1/media-imports accepts public TikTok or Instagram video or reel URLs only (YouTube and other hosts return unsupported_platform), at $0.15 per import. There is no documented upload endpoint for a local voice memo, so host the file yourself on any storage that serves it over public HTTPS.

Runnable example

Pass the URL and the memo length in seconds, so the reservation matches the audio.

import hashlib, os, sys, time, requests

API = "https://api.sume.com"
AUTH = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
url, seconds = sys.argv[1], int(sys.argv[2])
if seconds > 600:
    sys.exit("STT takes at most 600 s per request; split the memo first")
key = hashlib.sha256(f"{url}{seconds}".encode()).hexdigest()[:32]
r = requests.post(f"{API}/v1/stt-1.0/transcribe",
                  headers={**AUTH, "Idempotency-Key": key},
                  json={"audio_url": url, "duration_seconds": seconds})
r.raise_for_status()
job = r.json()["data"]
while True:
    s = requests.get(job["status_url"], headers=AUTH).json()["data"]
    if s["terminal"]:
        break
    time.sleep(s.get("next_poll_after_seconds") or 3)
res = requests.get(job["result_url"], headers=AUTH).json()["data"]["result"]
print(res["text"])

What if the memo is longer than 10 minutes?

Split it into parts of 10 minutes or less, transcribe each, and add each part's start time to its word timestamps. A 25-minute memo is three requests.