Three free voice apps, and the job each one actually wins

Before you spend money on a recorder, spend an afternoon with the free ones. We read the source of three of them, not the marketing pages, and each does a job better than we do.

Two of them are dictation tools and one is a self-hosted server, so none of the three is a meeting archive. That distinction is most of this post. If the job you actually have is one of theirs, use theirs.

Handy, if you want to talk instead of type

Handy is free and open source. You hold a key, you talk, and the text appears in whatever field your cursor is in. Email, chat, a code editor, a form. It never opens a window in your face.

Two things make it worth the download. It knows the names you use. Say "Charge B" and it writes "ChargeBee", because it matches on how a word sounds rather than how it is spelled, and it keeps your capital letters and punctuation intact. It also strips the phantom lines that speech models invent over silence, the "thank you for watching" class of nonsense that a raw transcript is full of.

It lets you pick from eight different speech engines and switch between them. That is more choice than most paid apps offer. Handy is a dictation tool and not a meeting archive, so judge it on how well it types for you.

Get Handy if: you type a lot and would rather talk. Skip it if: you need to keep and search recordings later.

Verenu, if the same word keeps coming out wrong

Verenu is free, open source, and its best feature is one nobody advertises. It watches the corrections you make. Fix the same mistranscribed word enough times and it learns the fix and stops asking. Reject the fix and it forgets it again. The whole memory is four columns: the mistake, the correction, how confident it is, how often it happened.

For a name, an acronym, or a product word that every transcriber butchers, that is the feature you actually want.

Get Verenu if: your work is full of jargon a general model has never heard.

Speakr, if you already run a home server

Speakr is a self-hosted server rather than a desktop app, so it suits people who already have a machine running at home and want transcription as one more service on it. Its templates are real. You can shape how a recording gets named, how it gets exported, and what the model is told before it starts, all separately.

Its source lives at github.com/murtaza-nasir/speakr, with a companion GPU service at whisperx-asr-service.

Get Speakr if: you self-host things already and want this to be one of them.

What we took from them

We say this plainly because these projects gave their work away and we used it. Speakr's split between "words to expect" and "context to give the model" is now how yapyap works, because one setting for both jobs was worse. Handy's filter for invented lines is an idea we took directly.

Where we differ is corrections. Verenu earns its learning through accessibility permissions that read what you edited in other applications. Our transcript editor is inside yapyap, so we see the same edit without watching anything outside our own window. Verenu's version reaches further than ours. Ours asks for less.

Where yapyap earns the money

None of the three records your phone and hands the file to your desktop over your own network. None of them keeps your microphone and the far end of a call on separate tracks so speaker labels cannot cross between them. And none of them lets one recording become minutes, an action list and a quote sheet from write-ups other people wrote and shared.

That is the line. If your problem is getting words into a text field, Handy is free and it is right there. If your problem is a year of conversations you need to search, label by who spoke, and turn into a different document for each person who asks, none of the three does that, and that is what yapyap is.