For most of last year, Trilla could tell you what was said. It just couldn't reliably tell you who said it. That bothered me, because half of why you record a meeting is to remember who committed to what.
The 1.11 release line is the one where I finally fixed that, and then kept pulling the thread until the whole experience after a meeting felt different.
Voices that follow you across meetings
When you finish your first recording in the new build, Trilla asks you a small question: which speaker is you? That single tap turns into a voice profile that lives on your machine. From then on, you don't show up as "Speaker 2" anymore. You show up as you.
Name a colleague once and Trilla recognizes them on the next call. The voice prints never leave your device — they sit in the same local database as your transcripts, encrypted with the rest of your data. There is no cloud directory of voiceprints. There is no "voice ID account." There is just a file on your laptop that Trilla reads when it needs to.
I wanted speaker recognition to work the way names work in real life. You meet someone, you remember them, you greet them by name next time. The app shouldn't need a server to do that.
Diarization that actually agrees with itself
Under the hood, Trilla also got a new offline diarization pipeline. The old one was solid but had a habit of swapping speaker IDs halfway through a long call — Speaker 1 in the first half would mysteriously become Speaker 3 in the second.
The new pipeline keeps speaker assignments stable across the whole recording, and the reprocess flow now matches new segments against your existing voice profiles instead of starting from scratch. If you've ever relabeled the same person three times in one transcript, this is the fix.
The dashboard nobody asked for, and now everyone uses
The other big change is what you see when you open the app on a Monday. Trilla now opens to a briefing: what happened across your meetings last week, the decisions you made, the action items still open, who you spent the most time talking to.
Behind that page is a small but real change to how Trilla treats your transcripts. After every recording it now extracts structured insights — decisions, action items, and topics — and stores them next to the transcript. The dashboard is just a window onto that data. Sparkline of meeting volume. Heatmap of when you actually take calls. Distribution of who's been on the other end of them.
I almost cut this feature twice. It felt indulgent. The thing that convinced me to keep it was opening it on my own machine after a busy week and realizing I'd forgotten about three commitments I'd made on Tuesday. Tools that make me less embarrassed are tools I want to use.
Smaller things that matter
A few rough edges that have been bothering me for a while are also gone. The recording stop handler no longer has the crash that some of you hit on very long calls. The waveform stays smooth even on multi-hour recordings. Speaker labels survive a stop-and-resume. Dashboard data refreshes the moment a recording finishes instead of waiting for a reload.
None of this is glamorous. All of it makes the app feel less like a beta and more like a thing you can rely on every day.
What's next
Importing recordings you already have. Pulling audio out of video files. A friendlier first-run. Most of that is already in the next release.
If you've been holding off on Trilla because the speaker labeling wasn't quite there, this is the build to try.