Weesper Neon Flow

Hi Fazier, Cyril here, one of the makers. Weesper Neon Flow is voice dictation for Mac and Windows: hold a shortcut, talk, let go, and the text shows up where your cursor is, in pretty much any app. Speech recognition runs locally with Whisper, so it keeps working offline once the model is installed. There's a 15-day free trial with no card, then a free plan capped at 1,500 words a week. I'd like to hear which apps you would dictate into first, and where transcription trips up for you.

Automatic language detection is useful for people who switch languages during the day. How does it handle a sentence that mixes two, like English technical terms inside Portuguese or Russian speech? Does the custom dictionary help there, or does Whisper settle on one language per recording?

Running Whisper locally so dictation works offline and audio never leaves the machine is a strong point for people who write sensitive emails. How much RAM and disk space does the default model need, and is there a smaller model for older laptops? I'd also like to know how the custom dictionary handles names that Whisper often mishears.

Hold-to-talk feels natural, but it's the same gesture as push-to-talk in Discord or Teams. Can the shortcut be a mouse button or a single key, and how does it avoid clashing with an app that's already listening for the same key? On the formatting rules: can I teach it that saying "slash remind" should type "/remind"? That would make dictating into chat apps with slash commands much smoother.

Offline dictation is a big deal for anything personal. We build a journaling app, and privacy is the first thing our users ask about. Does Neon Flow handle mixed-language speech, like English with Malay or Chinese words in the same sentence? That's very common for our users in Southeast Asia.

Running Whisper locally is a good fit for anyone dictating private emails or notes. Two things I'd want to know before installing: which model size does the default install use (and is there a smaller one for older Macs), and does hold-to-talk still work in terminals or remote desktop windows where the shortcut may be captured by the app itself?

Answering your question from the speech side: where transcription trips up for me is language switching. Bilingual people often change language mid-sentence, and detecting the language once per utterance picks one and garbles the rest. Can I pin a language to a shortcut, or does it detect per dictation? Also curious which Whisper model size ships by default and how long a 30-second dictation takes on an older Intel Mac. First app I'd dictate into: Slack.