Kroko Onnx

Kroko is a speech to text service built on compact commercial models designed to run on ordinary CPUs rather than GPUs. It handles both streaming and pre recorded transcription, with streaming partial results at about 320 milliseconds. The vendor states that it runs about 10 times faster than real time per CPU core and costs 5 to 20 times less than popular cloud speech APIs. It can be deployed three ways. The Cloud API is hosted in the vendor's European cloud and offers one API for streaming and recorded audio, WebSocket access, and word and segment timestamps. The On Premise option runs the commercial models on the customer's own infrastructure, keeps audio and transcripts in the customer's network, and lists deployment options for HIPAA and GDPR workloads and OEM licensing. The On Device SDK embeds the models in iOS, Android and browser apps through WASM and works offline. The vendor reports second place of nine on the Open ASR benchmark for average word error rate. Pricing is by audio hour: the Cloud API and On Premise options are listed at $0.05 per hour, with volume pricing down to $0.02 per hour. The On Device SDK starts from $150 per month, licensed by app scale. A free trial is offered.