Field notes on voice infrastructure.
Pricing breakdowns with sources. Public benchmarks, including the calls we lose. Engineering decisions from the team building the cheapest minute on the market.
Not the most accurate. The best accuracy per dollar for Latin American Spanish.
Fourth of seven on real Latin American calls, three points off the leader, at a fraction of the price. The math, the features that ship today, and how we size concurrency per customer instead of per plan.
WER vs DER vs SA-WER: the three speech metrics, and why two of them hide the failure that matters
Word error rate measures words and is blind to speakers. Diarization error rate measures time and is blind to words. Neither tells you whether the right words landed on the right speaker — the number that actually predicts whether a diarized transcript is usable.
The best STT API for conversation QA in 2026 — an independent benchmark
Five dimensions matter for picking an STT API under a conversation QA workload: price per minute, accuracy on regional accents, diarization, real-time latency, and multilingual coverage. Source-linked rankings across seven vendors, plus the one-line drop-in for the winner.
From AssemblyAI to Orchard in 60 seconds: a migration walkthrough
If you're running AssemblyAI in production, swapping to Orchard takes one configuration change, one response field rename, and a verification pass. The full mechanical walkthrough, with the cost delta on a 5M-minute workload.
Build vs buy: should your conversation QA platform run its own STT?
When the STT bill crosses six figures a month, every CTO is asked the same question by the board: why don't we bring this in-house? The honest decision tree, the hidden costs of self-hosting Whisper, and the case for buying from a vendor priced like infrastructure.
The diarization tax: the silent line item killing call center analytics margins
Speaker diarization is the feature that turns transcription into analytics. Most STT vendors price it as an optional add-on at a 30-50% uplift on base. Why it should be bundled, and what the math looks like when it is.
Why Gong, Observe.AI and CallMiner are still paying yesterday's STT prices
The three biggest conversation intelligence platforms run on STT pricing that hasn't moved with the market. A modeled estimate of what their bills look like at scale — and what changes for the next platform that doesn't carry their incumbency cost.
The 10M-minute audit: what speech analytics platforms actually pay for STT
If your platform processes 10 million minutes a month, your STT line item ranges from $4,200 to $240,000 depending on the vendor. The full per-minute audit with sourced prices, and the line items inside the line item.
Speech analytics in Spanish: why generic STT fails on Latin American calls
Generic Spanish STT is trained on Castilian and dubbed-neutral voice — neither of which is what your call center records. The phonetics behind the failure mode and the accent-embedded approach that fixes it.
How to cut conversation QA costs 90% without changing one line of code
If your QA platform runs on the OpenAI Whisper API or any Whisper-compatible provider, the cheapest STT minute on the market is a one-line migration away. The math, the swap, and what stays the same.
The hidden STT tax breaking conversation intelligence margins
Conversation intelligence looks like SaaS on the pitch deck and bleeds like usage-priced infra on the P&L. The math behind the floor most CI buyers never benchmark — and the API layer that lifts it.
Real-time STT for call coaching: under 200ms or your CSAT dies
Agent-assist platforms live and die by the gap between the customer's last syllable and the suggestion on the agent's screen. The cognitive science behind the 200ms threshold, where real-time STT vendors actually land, and the latency budget that fits inside a single conversational turn.
The cheapest speech analytics API in 2026: WER, RTF and the full per-minute breakdown
Speech analytics — transcription + speaker diarization + intelligence — should cost a fraction of a cent per minute in 2026. Orchard runs the full pipeline at $0.0030/min at production volume. Source-linked comparison across 12 providers, plus a frank read on WER and RTF.
The cheapest speech-to-text API in 2026: a 12-provider breakdown
Orchard's batch tier runs at $0.00042/min — 7× cheaper than Deepgram, 14× cheaper than AssemblyAI, 57× cheaper than AWS Transcribe. Source-linked comparison across 12 providers.