BENCHMARK GUIDE 01 · WINDOWS 10/11

Windows Local Voice Typing: Comparing Supersey 0.9.2 with Built-In Win+H

Windows 10/11 includes Win+H voice typing, which relies on cloud speech endpoints and frequently drops out offline, on restricted corporate networks, or during mixed-language technical dictation. Supersey for Windows (0.9.2) runs speech recognition and translation locally on your NVIDIA GPU (CUDA) or CPU — hold Right Alt to speak, release to insert text at your cursor.

Baseline version: Supersey Windows 0.9.2 · 2026-11-08 · Fact Sheet →

Supersey for Windows vs. Built-In Windows Voice Typing (Win + H)

DimensionSupersey for Windows (0.9.2)Windows Built-In Voice Typing (Win + H)
Execution & privacyRuns 100% locally on GPU (CUDA) or CPU; audio never leaves the PCRelies on cloud speech services; requires active internet connection
Trigger & pacingHold Right Alt and release to insert, plus optional smart-pause auto-commitPress Win+H to open floating bar; pauses often auto-mute the microphone
Mixed languages & jargonHandles mixed Chinese/English naturally; supports custom glossary and 'wrong=right' rulesSingle-language mode often mistranscribes English acronyms or code identifiers
Cursor injection compatibility3 modes (keystroke, paste, chunked typewriter) working across IDEs, terminals, and legacy appsRelies on Windows TSF text services; fails in some remote desktops and non-standard inputs
Direct spoken translationSpeak in your native language and insert translated English/Japanese/etc. directly at the cursorOutputs same-language transcription only

Key Windows Findings

Dual Local Path: NVIDIA CUDA + Multi-Core CPU

On desktops and laptops with an NVIDIA GPU (6 GB+ VRAM), transcription completes within hundreds of milliseconds after releasing Right Alt; laptops without a discrete GPU automatically run on CPU.

Windows Smart-Pause Continuous Mode

Alongside hold-to-talk, the Windows build supports smart pause detection that segments and commits long spoken paragraphs at natural breath pauses.

Three Injection Modes for IDEs & Terminals

Switching between simulated keystrokes, clipboard paste, and typewriter mode prevents dropped characters in VS Code, Windows Terminal, and enterprise desktop software.

Reproducible 5-Step Test Protocol

Use the following 5 steps to verify latency, accuracy, and offline operation on your own Windows 10/11 PC:

  1. 1

    Step 1: Record Hardware Baseline & Cache Local Weights

    Check Task Manager for your CPU model, RAM, and GPU VRAM. Install Supersey-Setup-0.9.2.exe and complete the one-time local weight download.

  2. 2

    Step 2: Enable Airplane Mode to Verify Offline Dictation

    Unplug Ethernet and turn on Windows Airplane Mode. Press Win+H to observe its network error, then hold Right Alt with Supersey and dictate into Notepad to verify offline insertion.

  3. 3

    Step 3: Run 3 Standardized 10-Sentence Prompts

    Dictate 10 sentences each of: (a) everyday conversational prose, (b) mixed-language technical notes containing terms like API, JSON, GitHub, and React, and (c) quiet speech through a Bluetooth headset 40 cm from the mic.

  4. 4

    Step 4: Test Deterministic Glossary Rules ('wrong=right')

    Take any proper noun or product name misheard in Step 3, add a 'misheard=correct' entry to the Supersey glossary, and repeat the sentence to confirm immediate correction.

  5. 5

    Step 5: Measure Key-Release Latency & Memory Footprint

    Dictate a 15-second paragraph (~60–80 words), time the interval from releasing Right Alt to text appearing at the cursor, and inspect RAM/VRAM usage in Task Manager.

Windows Local Voice Typing FAQ

OFFICIAL FACT SHEET · SINGLE SOURCE OF TRUTH

Fact Sheet

This page serves as the canonical reference for objective technical specifications, offline vs. network boundaries, hardware requirements, and one-time pricing across Supersey for macOS and Windows.

BENCHMARK GUIDE 02 · MACOS (APPLE SILICON)

Mac Local Voice Typing Benchmark

Apple Silicon Macs (M1 / M2 / M3 / M4) feature unified memory well suited for local speech and translation inference. Supersey for macOS (0.9.5) lets you hold Right ⌘ to dictate or translate directly into any cursor position, and pairs Apple's native Vision framework with local translation for on-screen OCR.

BENCHMARK GUIDE 03 · PRICING & ARCHITECTURE

One-Time Purchase vs. Subscription

Most modern AI dictation tools charge $12–$15 per month ($144–$180 per year) because every spoken sentence is streamed to cloud GPUs. Supersey runs speech recognition and translation on your own Mac or Windows hardware, enabling a straightforward one-time purchase ($15 Basic / $25 Plus / $35 Complete) with zero recurring fees.

SUPERSEY · macOS & WINDOWS

Just say it.

The next message, the next note, the next page you cannot read.