Skip to main content

Release Highlights

Latest

May 1, 2026

Channel-based Host/Guest speaker tags, dramatically improved Bluetooth headset reliability, faster Whisper performance, and a working GPU path for Windows.

+Host vs Guest speaker tags from mic and system audio+Bluetooth headset disconnects no longer wreck a recording+Switch audio devices mid-recording without losing the meeting+Faster transcription drain when you stop recording+Reliable GPU acceleration on Windows+Card-based AI provider selection across the app

What's New in Pro 1.7.3

Five weeks since 1.5.0, 244 commits, 12 PRs merged. Three things you'll notice immediately: Windows users with a GPU finally get clean accelerated transcription, Bluetooth headsets no longer kill a recording on disconnect, and your transcripts now carry speaker tags so you can see which voice is yours.


Host vs Guest Speaker Identification

Your transcripts now tag your voice as Host and the meeting audio as Guest, so it's clear who said what.

  • Deterministic tags from the audio source - Mic = Host, system audio = Guest. No more guessing.
  • Per-channel recording files - Each meeting saves a mixed recording plus separate mic and system audio tracks.
  • Inline editing - A pencil icon next to any speaker tag opens a quick editor right in the transcript. No need to dig through Settings.
  • Graceful degraded mode - If one audio source has a problem mid-meeting, only that side falls back to "Unidentified." The other side keeps its label, and a small indicator surfaces what happened.

Per-voice speaker identification - telling Alice from Bob inside the same channel - is on the roadmap and not part of this release.


Bluetooth Audio Survives

The biggest source of broken recordings was a Bluetooth headset disconnecting mid-meeting. That's now handled.

  • Automatic fallback when a Bluetooth mic disconnects, so the recording continues without a long silent gap.
  • Live device hot-swap - Switch your input device while recording, and the app keeps going.
  • Cleaner Bluetooth audio path that avoids the resampling artifacts BT headsets used to introduce.
  • Smarter echo handling on headphones - Prevents external-mic leakage from being mislabeled.
  • Resilient capture under load - Slow consumers no longer take down the recording stream.

Faster, More Stable Transcription

A focused 65-commit effort that cleaned up the most-reported transcription problems.

  • Dramatically faster drain when you stop a recording. What used to take tens of seconds now finishes in a few.
  • No more repeated-phrase hallucinations carried across chunks. The transcript stays clean even on long meetings.
  • Better non-English handling. Doubled output on CJK and other languages is fixed, and the engine no longer biases toward English when the audio isn't.
  • Per-recording language lock. Once detected, the language stays locked for the session. No mid-call flip-flop.
  • Quieter language indicator. Surfaces only when detection is meaningful.

Windows GPU Acceleration That Works

If you tried GPU acceleration on Windows in 1.5.0 and went back to CPU because of garbled transcripts, this release fixes it.

  • Clean transcripts on Windows with NVIDIA GPUs.
  • Short utterances like "hi" or "yes" now transcribe correctly without getting stuck in repetition loops.
  • GPU acceleration is still active - only the unstable code path was bypassed.

Card-Based AI Provider Selection

Every place you pick an AI model now uses the same card-based UI - in Settings, in the meeting view, and in preferences.

  • Configure before you switch. Picking an API provider opens a config dialog first; the provider doesn't activate until your key works.
  • Auto-fetch models as soon as a valid key is entered.
  • No stale errors when you switch providers.

Providers covered: Built-in AI, Hosted, Custom, Ollama, Claude, OpenAI, Groq, OpenRouter.


Other Improvements

  • Sidebar meeting metadata - Each meeting in the sidebar now shows its date and time below the title.
  • Speaker label editor moved to the transcript itself (pencil icon) so you can edit labels in context instead of opening Settings.
  • PDF and DOCX export work from any tab, not just the summary tab.
  • Toasts moved to the bottom-right to stop overlapping with action bars.
  • Recording timer pill in the meeting top bar lets you jump back to the recording screen with one click.
  • Retranscribed segments no longer show misleading generic "Speaker" tags.

Upgrade

The app will prompt you to update via the in-app update dialog. Click "Update Now" to install. All existing settings, recordings, and models are preserved. The transition from 1.5.0 is automatic.


Download

Or visit the downloads page for guided install with automatic platform detection.

Get started

Ready to try Meetily?

Join 180,000+ users who use Meetily for private meeting transcription. No bots, privacy first. Community Edition free.

No meeting bots
100% local transcription
Free & open source
Download Free

Star on GitHub (11.6K+) · Open source & self-hostable