Last updated: 2026-06-22

Audio transcription cost explained

The audio transcription cost of a speech-to-text API is the price charged to convert a recording into written text, computed from the length of the audio and the provider's published per-minute rate. Anyone budgeting for podcast transcripts, meeting notes, call-centre logs, or subtitle generation needs this figure, because hosted transcription is metered by how long the audio runs, not by how many words it contains.

How transcription is billed

Speech-to-text providers quote a rate per minute of source audio. The number of words spoken in that minute does not change the bill — a minute of dense dialogue and a minute of slow narration cost the same. Partial minutes are commonly rounded up, so a recording of three minutes and ten seconds is billed as four minutes by some providers. The model here uses the exact duration; if your provider rounds, round the duration up before entering it.

The formula

With a recording of duration $t$ minutes and a price $p$ per minute, the cost is the product of the two:

When the duration is given in another time unit, it is first converted to minutes, since the rate is quoted per minute. The relationship is linear: each additional minute of audio adds exactly the per-minute rate to the bill.

Worked example

Consider a 47-minute interview transcribed at a rate of $0.006 per minute, a common hosted speech-to-text tier. The cost is

so the full interview costs a little over a quarter of a dollar. A weekly podcast of this length would cost roughly $15 per year to transcribe — small enough that the duration, not the rate, is usually the figure worth checking.

Notes and variations

Optional features change the effective rate. Speaker diarization, word-level timestamps, and language identification are sometimes billed at a higher per-minute rate or as separate add-ons, so set the rate to the tier you actually use. Stereo recordings with multiple channels can be billed per channel; down-mixing to mono before upload avoids paying twice for the same content. Trimming leading and trailing silence reduces the billed duration directly.

Application

Per-recording cost is the building block for larger transcription budgets. Multiplying it by the number of recordings per month gives recurring spend. For the reverse direction — synthesizing speech from text rather than transcribing it — see the Text-to-Speech Cost Calculator. When transcription is one line item in a broader AI workload, fold it into the LLM Monthly Cost Calculator.