Performance Benchmarks

Real production measurements, not marketing rounding. Lokutor runs the entire voice pipeline on a single commodity CPU server — here's exactly where the time and cost goes, with the methodology to reproduce it.

The Full Pipeline Latency

In real-world voice AI, TTS latency alone doesn't tell the story. We measure the end-to-end (E2E) cycle: from the moment the user stops speaking to the moment our agent starts responding.

Where the time goes (single session)

Time-to-first-audio breakdown on our production stack (AWS m7g.xlarge, Graviton4, 4 vCPU / 16 GB ARM64, K3s, ONNX Runtime).

The STT and LLM stages run on best-in-class third-party low-latency providers, not our own hardware — over half the total is network and cloud-LLM time outside our pipeline. This is one representative single-session run.

Cost per unit (production pricing)

Where CPU-native inference actually shows up: cost, not a latency leaderboard.

Published competitor pricing is public list pricing; ours is measured production cost on our own infrastructure.

Intelligent Noise Suppression

Psst, our proprietary spectral-gating noise suppression model, runs as low-latency middleware ahead of STT to isolate speech in difficult acoustic environments (cafes, wind, traffic). We're currently running formal held-out evaluations for word-error-rate impact by environment — quantified results will be published here once complete, alongside our TTS quality (MOS) and speaker-similarity sweeps for Versa 2.0.

Pricing & Scale

Lokutor's entire production stack runs on a single CPU node — no GPU fleet to provision, no idle-capacity bill. That efficiency shows up directly in what you pay.

Why CPU-Native Matters

State-of-the-art voice AI today means giant models on giant GPU fleets. That model breaks down the moment voice AI needs to live outside a data center.

• Lifetime cost: sell a voice-enabled device once, and GPU-dependent inference means paying cloud bills for that device's entire lifetime. CPU-native inference removes that tax.
• Latency & connectivity: round-trips to a GPU cloud fight against real-time conversation. Running on the CPU already in the device (or a commodity server) removes a hop.
• Privacy & compliance: for regulated buyers, audio that never has to leave the customer's environment shortens procurement under GDPR and the EU AI Act — compliance as a distribution advantage, not a cost center.

Verify Data

Want to run these benchmarks on your own data? Contact us for a technical consultation or access to our evaluation scripts.