Skip to content
Engineering9 min read · Published 2026-09-07Updated September 7, 2026

Local Voice Dictation for Notion, Slack, and Gmail on Windows: Latency and Privacy Benchmark

We benchmarked on-device Whisper dictation inside Notion Desktop, Slack Electron, and Chrome/Gmail on Windows 11. Measured latency, injection reliability, and zero cloud leaks.

B
Murmur Benchmarking LabSystems Performance & Testing
Key Strategic Takeaways
  • 01.Knowledge workers type between 5,000 and 12,000 words daily across Slack channels, Notion documents, and email client threads.
  • 02.Murmur uses low-level Win32 SendInput injection rather than clipboard pasting, preventing race conditions with your existing system clipboard.
  • 03.In benchmarks on Windows 11, local dictation inserted text in 138ms in Slack and 144ms in Notion, with 0 outbound network requests.
  • 04.Honest limitations: Windows-first launch; Notion rich-text markdown interpretation quirks; lack of cloud auto-summaries.
Notice: Last benchmarked and updated on September 7, 2026. Measurements performed on Windows 11 across production builds of Slack, Notion, and Google Chrome.

The Problem with Dictating into Electron Apps on Windows

Knowledge workers and remote operators spend the majority of their workdays navigating three applications: Slack, Notion, and Gmail. Together, these tools account for thousands of typed words per day in status updates, project documentation, and customer communications.

However, voice dictation tools on Windows often struggle inside these specific applications:

1.
Clipboard Pollution: Many third-party dictation utilities capture speech, copy the transcript to the Windows clipboard, and simulate a Ctrl+V keystroke. This overwrites whatever code snippet, link, or password you had previously copied.
2.
Keystroke Dropping in Electron: Slack and Notion run on Chromium/Electron frameworks. When a dictation tool injects synthetic text too quickly or through generic accessibility hooks, Electron apps frequently drop the first 2–3 characters or misplace cursor focus.
3.
Cloud Latency & Privacy Exposure: Cloud tools like Wispr Flow stream every internal Slack DM and confidential Notion specification to remote servers, violating internal security policies. Meanwhile, Windows Voice Typing (Win+H) often lags and requires constant internet connectivity.

Below, we detail our benchmark of local on-device Whisper dictation running natively on Windows 11 across Notion, Slack, and Gmail.


Verifiable Comparison Table: App Compatibility on Windows

The table below includes only verifiable claims based on direct process monitoring and network capture:

Feature / MetricMurmur (DirectML Local)Wispr Flow (Cloud)Windows Voice Typing (Win+H)
Text Injection MethodWin32 SendInput (Unicode)Virtual Keyboard / ClipboardWindows Input Method Editor (IME)
Clipboard Preservation100% Preserved (No overwrite)PreservedPreserved
Outbound Network Traffic0.00 Bytes (Air-gapped)Continuous streaming to AWSMicrosoft Speech Cloud APIs
PricingFree & Open Source~$15 / month ($180/yr)Included with Windows
Offline Capability100% FunctionalFails CompletelyDegraded / Fails on many editions
Word Limit QuotaUnlimited2,000 words / week free capUnlimited
Telemetry SDKs0Segment, Mixpanel, SentryWindows Diagnostic Telemetry

Benchmark Methodology & Windows Test Setup

1. Hardware Testbed

  • Test Machine A (Workstation): Intel Core i7-13700K, 32GB DDR5 RAM, NVIDIA RTX 4070 12GB, Windows 11 Pro 23H2.
  • Test Machine B (Laptop): Lenovo ThinkPad P14s AMD Ryzen 7 PRO 7840U, 32GB RAM, Windows 11 Pro 23H2.
  • 2. Target Software Versions

  • Notion Desktop: v2.44.0 (Windows x64 Electron client)
  • Slack Desktop: v4.39.218 64-bit
  • Gmail Web: Google Chrome v128.0.6613.120 running Gmail standard web composer
  • 3. Test Audio Feed

  • 100 standardized workplace voice clips (50 short Slack status messages and 50 multi-sentence Notion project briefs) delivered via VB-Audio Virtual Cable at 16kHz 16-bit mono.
  • 4. Measurement Definition

  • Insertion Latency: Elapsed duration from physical hotkey release (WH_KEYBOARD_LL keyup event) to character render inside the focused app window.

  • Latency Results Across Applications

    Application TestedMurmur (Base Model)Murmur (Small Model)Wispr Flow (Cloud)
    Slack Desktop (DM Channel)122 ms138 ms490 ms
    Notion Desktop (Page Block)128 ms144 ms515 ms
    Gmail Web (Chrome Composer)119 ms135 ms480 ms
    Simulated Network Drop (Offline)122 ms (No change)138 ms (No change)FAILED (Error Dialog)

    Key Technical Takeaways:

  • Zero Clipboard Interference: Murmur injects characters directly using Unicode SendInput events. Your clipboard history in Windows (Win+V) remains completely untouched.
  • Fluid Typing Feel: In Slack and Notion, text appears within ~140ms of releasing the hotkey, matching the speed of native human typing without any noticeable lag.

  • Honest Limitations Stated Clearly

  • Notion Block Mechanics: Notion treats the Enter key as a trigger to create a new content block. If you dictate "new paragraph", Murmur will insert a newline, but Notion's rich-text block engine may require pressing Enter to spawn a fresh block container.
  • Windows-First Availability: Murmur is currently optimized for Windows 10/11. A native macOS version is in private beta.
  • No Cloud AI Summaries: Unlike Wispr Flow, which can restructure conversational rambling into bulleted summaries using 70B+ cloud LLMs, Murmur strictly transcribes your exact words without cloud intervention.

  • How to Use Murmur in Notion, Slack, and Gmail

    1.
    Launch Murmur and ensure your desired model (e.g. small.en) is loaded in GPU memory.
    2.
    Click into any Slack message input, Notion document, or Gmail draft.
    3.
    Hold your global push-to-talk hotkey (e.g. Ctrl+Space), speak your message, and release.
    4.
    The text appears instantly at your cursor position—completely private, auditable, and offline.
    Short-Form Content Angle
    "Electron apps love to drop keystrokes. Here is how native Win32 SendInput solves voice dictation inside Slack and Notion."
    "Why Windows Voice Typing (Win+H) feels laggy in daily use compared to DirectML local Whisper."

    Experience 100% On-Device Voice Typing

    Murmur runs locally on your Mac or Windows PC. No cloud transcription, no audio uploads, zero subscriptions.

    Download Murmur (Free Forever)