Per-hour streaming rate vs per-minute STT: how a short clip rounds

Microsoft's announcement lists MAI-Transcribe-2-Streaming at $0.54 per hour through the end of the year, and the page does not say how a session shorter than an hour is rounded, so do not assume per-second proration. Sume STT 1.0 is billed per audio minute rounded up from your duration_seconds hint, and it reserves one minute when you omit the hint, so a 45 second clip is one minute and a 61 second clip is two.

What each side states

On the Microsoft side, the announcement gives the hourly figure and the intro window. It names no minimum, no rounding step and no billing granularity, so a bill for short sessions is something to measure rather than compute.

On the Sume side, the pricing code states that only transcription bills: per ceil of the audio minutes in the hint, one minute when the hint is absent, with a maximum of 10 minutes. The public rate is $0.01 per audio minute.

Compute the Sume side exactly

Use integers for money. This function turns a duration hint into billed minutes and micro-dollars, with the 10 minute cap and the one-minute default.

import math

RATE_MICROS = 10_000  # $0.01 per audio minute

def billed(seconds=None):
    if seconds is None:
        minutes = 1
    else:
        if not 1 <= seconds <= 600:
            raise ValueError("duration_seconds must be 1-600")
        minutes = math.ceil(seconds / 60)
    return minutes, minutes * RATE_MICROS

for s in (None, 45, 60, 61, 600):
    print(s, billed(s))

Limits

Rounding rarely matters for one file and matters a lot for thousands of short ones, which is why the hint should be your real length rather than the maximum. The point is that the two services are not billing the same unit, so a total comparison needs a measured sample of your own durations, not a single division.

Use this for your own planning

Write down the unit each vendor actually bills before you compare totals.

The Microsoft page gives a rate per hour and an intro window through the end of the year, so ask or measure how a one-minute session is charged. On Sume the unit is the minute, applied to your hint, so a good hint keeps the reservation close to the audio.

A measurement plan

Run ten sessions of different lengths, such as 20 seconds, one minute and five minutes, and compare the invoice line with your own timer. For Sume, compare the usage record for each job with the minutes in your hint. Differences between the two will tell you more than any list price, and they cost a few cents to find.