OpenASR Desktop 0.1.21: Voice ID
Enroll a voiceprint so transcripts can name who is speaking. With built-in MOSS speaker separation, multi-person meetings stay split — and named. Local-first.
OpenASR Desktop 0.1.21 is out. The headline is Voice ID: enroll a voiceprint once, and transcripts can name who is speaking — not only that someone new started talking.
It sits next to the built-in MOSS speaker separation already in the app. Together they turn a multi-person recording from a wall of text into a conversation you can follow.
Voice ID
Meetings, interviews, and long calls share a simple problem. Speaker labels like “Speaker 1” and “Speaker 2” keep turns apart, but they do not tell you who those people are. Voice ID closes that gap.
Enroll a person with a short clip of their speech — about ten seconds is often enough for enrollment to complete on its own. Common formats work: mp3, m4a, and aac, not only wav. If enrollment cannot finish, you get a clear reason instead of a silent failure. Names can be edited afterward, so a rough first pass does not lock you in.
In the marketplace, models that support Voice ID are labeled. Pick one that does, enroll the people who matter, and the next transcript can carry their names.
Who said what
Speaker separation is the other half. MOSS labels who spoke while the file is transcribed, so turns stay split across a meeting or interview even when you have not enrolled anyone.
With Voice ID, that split can stay tied to the same people across a long file and across segments — not a fresh “Speaker 3” every time the same voice returns. Separation keeps the conversation readable; enrollment makes it personal. Both stay on your machine by default. Audio does not leave the device unless you choose a path that does.
Also in 0.1.21
- Dictation — rapid re-press, or release and press again, no longer falsely reports busy or drops audio.
- More audio formats — including ADPCM wav. When a file is too large for available memory, capacity checks reject early with a breakdown (on Apple Silicon, swap is counted).
- Upgrades that keep your setup — installed models are preserved across the update; startup is stabler; default CUDA GPU coverage is broader, including Turing.
- New marketplace models — Fun-ASR-Nano and Granite Speech 4.1 2B.
Desktop 0.1.21 ships with engine 0.1.27.
Download
Get OpenASR Desktop from the download page, or take the direct links:
The speech engine underneath is the Apache-2.0 open core. Read it, run it from the CLI, or just open the app and talk.
Release 0.1.x in ten days: what fourteen versions built
Fourteen versions in ten days. The 0.1.x series was a foundation-laying sprint — halving resident memory, shipping 26 models across eleven platform targets, and hunting every fail-closed gap until idle unload actually worked. Here's what it built.
Release OpenASR Desktop is now on Windows
Same local-first speech-to-text, now on Windows. Same open core, same privacy boundary, same models. Here's what to expect — including the SmartScreen prompt you'll see on first launch.
Release OpenASR 0.1.0
The first public release: a local speech-to-text app for macOS over an Apache-2.0 open core. Here's what works today, what's still missing, and where to read the source.