Diagnose Vox latency issues via health checks, warm-up, benchmarks, and dashboard inspection.
npx skills add https://github.com/arach/vox --skill operations-guide-arach
Please help me install this Agent Skill. Skill: Operations Guide Source: https://github.com/arach/vox/tree/main/docs Command: npx skills add https://github.com/arach/vox --skill operations-guide-arach
This guide helps you diagnose and resolve Vox latency or reliability issues by turning vague “it feels slow” reports into actionable checks of warm-up, inference, and end-to-end runtime metrics.
Run the Vox operations loop by executing the doctor check first, then starting warm-up, benchmarking a real audio file, and finally opening the performance dashboard filtered to your integration’s clientId.
High-intent search queries and answers about installing and using this skill.
To triage Vox latency, run health checks and explicit warm-ups to separate cold-path modelLoadMs costs from hot-path inferenceMs and totalMs metrics. This isolates whether slow audio transcription or synthesis stems from initial model loading or active processing.
The best workflow for diagnosing cold versus warm Vox behavior starts with a doctor check, followed by model warm-up, benchmarking a real audio file, and inspecting the performance dashboard. This ordered runtime verification isolates latency shifts across transcription and synthesis workflows.
Interpret Vox telemetry by analyzing inferenceMs for active processing speed, totalMs for end-to-end runtime, and modelLoadMs for initial loading context. This metric interpretation clarifies whether latency spikes occur during model loading or actual audio inference.
Vox performance benchmarks become inconsistent if clientId, route, and modelId are not kept stable across tests. Contributor-safe guidance requires preserving these telemetry dimensions so performance comparisons remain meaningful over time across multiple client integrations.
Run explicit model warm-up before benchmarking audio processing whenever you need to measure hot-path inference latency without cold-start overhead. Warming models ensures repeatable benchmarks that reflect actual transcription and synthesis runtime rather than initial modelLoadMs delays.
Yes, Vox performance triage applies to diagnosing cold-vs-warm behavior across both transcription and synthesis workflows. Using consistent telemetry dimensions and ordered runtime verification steps ensures accurate latency diagnosis across multiple client integrations.
0xmariowuRun local deterministic health checks on the Builder Data Collection system.
ovachieverAutomate ML experiment tracking and model registry management via the wandb API.
heximaoGenerate audio spectrograms and analyze audio features from the command line.
ShmurdocDetect deploy platforms and write deployment settings to CLAUDE.md.
