← sources

how do you actually measure transcription cost per call, not per minute?

source: reddit (reddit:AI_Agents)published 2026-10-12 15:15ingested 2026-10-13 02:36

retained text

I keep seeing debates about which ASR model is "cheapest"! Nobody talks in per-call dollars. They only quote per-minute rates! Those numbers are hard to compare across vendors with different pricing tiers.

  
So I ran my own transcriber on a batch of recorded calls using Whisper. Transcribing a single call came out to about $0.02. That's it. Compared to voice engine costs and latency tuning, that number felt almost irrelevant.

  
It also caught something a realtime ASR missed! On one call the live transcript clipped a price down from "five fifty" to just "fifty." Whisper recovered the full $5.50 from the recording afterward. So there's an accuracy argument for running a second pass. It's not just a cost argument.

  
My question for people running their own pipelines: at what call volume does $0.02 per call start mattering in your budget. Is anyone picking a cheaper or faster ASR model purely to save on this line item, or is it always dwarfed by TTS and LLM costs. I'm trying to figure out if optimizing transcription cost is worth any engineering time at all. Maybe it's a rounding error until you're doing serious volume.

  
Curious what thresholds people have hit in practice. Not theoretical math, actual numbers from actual pipelines. Does anyone track cost per call instead of per minute already, and if so what does that log look like.

document fetched 2026-10-13 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1x44k5o/how_do_you_actually_measure_transcription_cost