Transcribe audio from a private bucket: Sume STT needs a public URL

You cannot point Sume STT at a private bucket path. The audio_url field must be a public HTTPS URL, and Sume's docs say signed or private URLs are rejected for media inputs. So the working pattern is to put the file somewhere Sume can fetch it over plain HTTPS, run the job, and take the file down afterwards.

Below is what the schema says, what to do about a private archive, and a short script that submits and polls.

What does the STT schema require?

In the Sume OpenAPI spec the STT 1.0 request has one required field, audio_url, described as a public HTTPS audio URL, with a preference for a Sume media or attachment URL. The asset library workflow states the same rule for other media inputs.

How do you hand over a private file?

Pick the least exposure that works for your data. None of these is something Sume does for you.

• Copy only the files you need to a separate public location, name them with random strings, and delete them when the job is terminal.
• Use a Sume media or attachment URL where your workflow already uploads through Sume.
• Do not publish anything you are not allowed to expose; if the audio is confidential, a public link is the wrong tool.

What does the submit-and-poll code look like?

The key is read from the environment. The job id comes back in data.job.id, and the status route is /v1/jobs/{id}/status.

import os, time, requests

H = {"Authorization": "Bearer " + os.environ["SUME_API_KEY"]}
B = "https://api.sume.com"

r = requests.post(B + "/v1/stt-1.0/transcribe", headers=H, json={
    "audio_url": os.environ["AUDIO_URL"],
    "language_code": "en",
}, timeout=60)
r.raise_for_status()
job_id = r.json()["data"]["job"]["id"]

while True:
    s = requests.get(B + f"/v1/jobs/{job_id}/status", headers=H, timeout=30)
    d = s.json()["data"]
    if d["terminal"]:
        break
    time.sleep(d.get("recommended_poll_interval_seconds", 3))

res = requests.get(B + f"/v1/jobs/{job_id}/result", headers=H, timeout=30)
print(d["sume_status"], res.status_code)

What does it cost?

STT is $0.01 per audio minute. Set duration_seconds (up to 600) when you know the length; if you omit it, the reservation is for one minute. For long archives see detaching and transcribing a podcast back catalog.