Blog post · November 19, 2026 · aviznetworks.com
Aviz Networks adds on-device voice to Network Copilot
- About Aviz Networks
- Also names Customers
- Data
- It
- Network
- Nvidia
- Read the original
Summary
Aviz Networks added Evokable to Network Copilot, a voice feature supporting 18 languages. Speech recognition and voice synthesis run on-device in the browser, keeping voice audio off Network Copilot servers, which matters for security teams.
Feature Overview
Evokable is available on every Network Copilot page through a microphone button. Users say the wake phrase, ask a question, and watch the transcript appear as they speak. The question is sent after a pause, the response is read back, and the system re-arms for the next query.
The microphone is inactive while Network Copilot is speaking, so the system does not capture its own output. Screen reader users can set a preference that silences Evokable, since the screen reader already provides audio feedback. Evokable is live in Chrome, Edge and Safari, and users enable it under Preferences, then Accessibility.
Language Support
Network Copilot supports 18 languages across the full interface, including labels, suggestion chips, system messages and settings pages, and the assistant can listen and respond in those languages. Japanese is the first non-Latin locale, which the company selected to test mixed scripts, unspaced text and full-width punctuation.
Language is set per user, with an Auto option that follows the browser, so teams in different regions can share one instance. The blog offers an example of an English-speaking engineer in North America and a Japanese-speaking engineer during an incident, with Network Copilot translating technical conversations while preserving terminology, configurations and remediation steps. Responses in other languages are translated before display, and on Chrome 138 and later that translation runs on-device.
Vocabulary and Customization
Evokable includes an organization pack with 34 pre-loaded networking terms, and users can add custom terms with both written forms and spoken pronunciations. Vocabulary can be exported, imported and shared across a team.
Wake, send and cancel phrases are configurable. By default, saying that's all sends a query, scratch that cancels it, and over functions as a radio-style end-of-transmission cue.
Data Handling
Speech recognition, transcript refinement and voice synthesis run in the browser using NVIDIA Parakeet and Kokoro models accelerated by WebGPU. According to the blog, voice audio is not sent to Network Copilot servers for processing, logging or any other purpose.
The microphone is open only while armed and cannot be reopened by the application after it is paused. While waiting for the wake phrase, the recognizer holds about two seconds of rolling audio, which is discarded and not stored, shown or sent. Evokable adds no widget, listeners or model downloads until voice is enabled, and the company reports no backend changes to Network Copilot server architecture.
Accessibility
Aviz states that Network Copilot voice was built to align with Section 508 from the outset. The company says it is completing a full WCAG 2.1 AA evaluation and will publish an Accessibility Conformance Report alongside the VPAT.
The company describes voice input and output, per-user language settings and screen-reader awareness as features that support engineers with different abilities and language needs. Customers that require procurement documentation can contact their account team for status details.
Evokable lets engineers query Network Copilot aloud in 18 languages, with voice processing kept in the browser and no backend changes reported. For on-call engineers, global operations teams and federal buyers, the blog presents the feature as addressing input, language and data exposure concerns within one workflow. This Blog Summary is a fact-based summary of the vendor blog and is intended to inform enterprise IT and security decision-makers.
Our summary. The original keeps its own byline.