Sume STT language_code: set a hint or omit it for auto-detect?

Pass language_code when you know the language of the recording, and omit it when you do not. On Sume STT the field is optional: the spec describes it as a BCP-47 or provider language hint, for example en or ko, and says to omit it for auto-detect. It takes one value. Sume does not publish a measured accuracy difference between hinting and auto-detect, so run a short sample of your own audio before you decide for a whole archive.

The practical rule is about what you know, not what you hope.

What does the spec say?

In the Sume OpenAPI spec, language_code is an optional hint with a single value, and omitting it asks for auto-detection. It sits beside duration_seconds, segmentation and metadata, none of which depend on it.

How do you test it cheaply?

STT is $0.01 per audio minute, so a comparison on a 2 minute clip with and without the hint costs about two cents each. Run both, read both, and keep whichever is closer to what you hear.

• Use a clip with names, numbers and fast speech, the hard parts.
• Compare transcripts by eye on the same ten sentences.
• Record which setting you chose in your own metadata.

What does the request look like?

Pass the hint from the environment, or leave it out:

import os, requests

H = {"Authorization": "Bearer " + os.environ["SUME_API_KEY"]}
body = {"audio_url": os.environ["AUDIO_URL"], "duration_seconds": 120}
if os.environ.get("LANG_HINT"):
    body["language_code"] = os.environ["LANG_HINT"]

r = requests.post("https://api.sume.com/v1/stt-1.0/transcribe",
                  headers=H, json=body, timeout=60)
r.raise_for_status()
print(r.json()["data"]["job"]["id"])

What about multi-language audio?

Sume takes a single hint, so see how another vendor handles several in multiple expected languages vs Sume's single language_code. For pricing across providers, see the cheapest speech-to-text API per hour.