← Back to the blog

Voice Typing Used to Be Bad. Here's What Actually Changed.

2 min read

Voice dictation has genuinely improved—modern speech recognition no longer produces the garbled mess that gave it a terrible reputation a decade ago. But the honest truth is that while it's now usable for real work, it still benefits from a quick edit pass, and how you use it matters more than the raw accuracy numbers.

Voice dictation has genuinely improved—modern speech recognition no longer produces the garbled mess that gave it a terrible reputation a decade ago. But the honest truth is that while it's now usable for real work, it still benefits from a quick edit pass, and how you use it matters more than the raw accuracy numbers.

---

Recognition accuracy

DictDrop: Modern speech-to-text catches the intent right the first time in most cases, letting you dictate code, prose, or commands without constant corrections. You'll spot the occasional mishearing—especially with technical terms or overlapping speech—but it's rare enough that a light review feels like polish, not recovery.

The general approach: Older systems genuinely struggled and required rewording sentences after the fact. Typical modern dictation is far better, but without active training for your vocabulary and domain, you'll hit more false starts than you'd expect.

---

How you activate it

DictDrop: Always-on with no hotkey, held button, or cursor pre-positioning—text appears as you speak. This removes the friction that made older voice typing feel like a separate mode; it's integrated into your actual workflow.

The general approach: Most built-in or typical voice dictation requires manual triggering (hotkey, button, or menu), which breaks rhythm and forces a context switch into 'dictation mode'—the same friction that made old voice typing feel clunky and inconvenient.

---

Privacy and audio handling

DictDrop: On Standard and Pro alike, you bring your own AI provider account: your audio goes straight from your laptop to that provider, and DictDrop has no server in the path.

The general approach: Older systems sent audio to a single cloud service with minimal transparency. Today's typical speech-to-text still assumes the cloud with limited control over retention or access logs—better than before, but privacy isn't the default.

---

Language support

DictDrop: Over a hundred languages with automatic per-utterance detection means you can code-switch or work across regions without reconfiguration. The old 'English or nothing' limitation is gone.

The general approach: Modern dictation supports major languages but typically requires manual selection upfront. Multi-language workflows or mixed-language utterances still require you to step in and change settings.

---