Sparrow-2
A turn-taking model that knows when a voice agent should speak
Emerged finds new products worth a look, daily — what’s new today
Lokutor
Voice agents that run entirely on CPUs, in the cloud, on-prem or on-device
A full voice-agent pipeline with no GPUs means cheaper hosting and a real on-prem or offline option for teams that can't send audio out.
Lokutor is a voice agent platform whose models run on ordinary CPUs. Its pipeline pairs its own noise suppression (Psst), turn-taking that tells a mhm from a real interruption (Turno), speech-to-text (Conv) and speech synthesis (Versa) with your chosen LLM. It claims about 120 ms to first audio and 0.9 s to first reply on one 4-vCPU node, across 33 languages and 10 voices with visemes. A free plan gives 30 minutes a month; paid plans start at $29, with on-prem and on-device licences quoted. Built by Hashing Works in Madrid.
More like this
Nari Qwen3-TTS and Qwen3-ASR
Fast, low-cost speech APIs built on Qwen3 voice models
Nico
Hold a key, say what you need, and a sandboxed worker on your Mac does it
Jinfer
An AI inference engine that runs inside the JVM, with no Python sidecar
Voxel
An open-source desktop voice changer that feeds any app through a virtual mic
Yogurt
Meeting notes that stay on your computer, with no bot joining the call