PetSpeak Blog

Notes from building PetSpeak.

How we train the models, the engineering decisions behind them, and the analysis we run along the way. Written for people who want the real details, not marketing.

What it costs to train our dog-sound model on cloud GPUs

A grounded cost analysis from a measured throughput benchmark and live RunPod pricing. Right-sizing the GPU beats raw horsepower — a full run costs about a third of a dollar.

July 2026Read post