Dictation on Hacker News: Mentions per Month

Developers and founders on Hacker News talk about dictation more than they did a year ago, and much more than a decade ago. The rise is not spread evenly across the field. The plain word "dictation" shows the clearest jump, with 682 mentions in the latest year ¹. The older phrase "speech recognition" is close to where it was, at 249 mentions ². The biggest gains belong to a handful of newer products and open models such as Wispr Flow, Voxtral and Parakeet.

This page covers 23 phrases ³. Each row counts the Hacker News stories and comments that contain one quoted phrase in the latest year, next to the count for the year before. The full table sits below this text. For background on the technology itself, see our overview of speech recognition.

The general vocabulary is growing, led by "dictation"

Four broad phrases set the baseline for how often the topic comes up.

"Transcription" is the biggest of them, with 1,627 mentions ⁴. It is also the least specific. The word covers meeting notes, medical records, podcasts and genetics, so a high count says little about dictation as a way of writing. "Speech to text" follows with 616 mentions ⁵, and it too sits above its prior-year figure.

"Dictation" has the steepest climb among the broad terms. Its latest count of 682 ¹ compares with a prior-year figure that is less than half as large, which you can check in the table. "Voice typing" is small, at 51 mentions ⁶, but it also more than doubled its prior-year count.

"Speech recognition" is the exception. Its latest count is only slightly above its prior-year figure, and the monthly series is far from its own record: the peak came in an early year, when the phrase was a standard label for any voice-driven product, and the most recent month is well below it. The phrase is still common, but it is not where the new conversation is happening. People now say "dictation" or name a tool.

Newer products and open models drive most of the growth

The sharpest rises belong to products and models that are recent, and several of them are open source or run on your own machine.

• Wispr Flow reached 137 mentions ⁷, several times its prior-year count. It is the largest named dictation app in the set.
• Voxtral reached 115 mentions ⁸, from a small base, and most of it came in a single burst around its release month.
• Deepgram, a speech API for developers, reached 127 mentions ⁹, about double its prior-year count.
• MacWhisper reached 87 mentions ¹⁰, and Superwhisper reached 76 ¹¹. Both are Mac apps built on Whisper, and both grew, MacWhisper faster.
• AssemblyAI reached 47 mentions ¹², a moderate gain.

The pattern is that attention follows launches. Open models and local apps get discussed in long threads, and a single popular thread can add a large share of a year's count. The peak months in the table show this: most products have one spike, not a steady climb.

Older and narrower products are flat or shrinking

Not every product grew. Several are at or below their prior-year counts.

• Otter.ai fell to 10 mentions ¹³, less than half its prior-year count. Its peak was years ago.
• Aqua Voice fell to 7 mentions ¹⁴, half its prior-year count, after a burst of interest at launch.
• gpt-4o-transcribe fell to 6 mentions ¹⁵, because its news moment passed.
• VoiceInk (21 mentions ¹⁶) and Speechmatics (11 mentions ¹⁷) are slightly below their prior-year figures.
• faster-whisper (40 mentions ¹⁸) and whisper.cpp (100 mentions ¹⁹) are close to level. These are tools developers build on, so their steady talk fits a mature layer under newer apps.

The legacy desktop products and niche tools stay tiny. Dragon NaturallySpeaking drew 8 mentions ²⁰, and Talon Voice, a hands-free control tool, drew 5 ²¹. Dragon rose from a very small base, so the gain says little. Hacker News readers are more likely to talk about the engine than about the established product.

Some of the biggest numbers are not about dictation at all

Three rows show why a phrase count needs care. Parakeet has 434 mentions ²², the most of any named model, and well above its prior-year count. It is also a bird. Most of the recent mentions refer to NVIDIA's speech model, but the search cannot tell the two apart. Moonshine has 116 mentions ²³, and it is also a word for illegal liquor. "Whisper" has the same problem, which is why the set uses longer phrases such as "whisper.cpp".

Even where the word is unambiguous, the figure is a count of mentions and not of users. A mention can be praise, a bug report, a complaint or a passing reference. One long thread can add hundreds in a month.

Where the sources disagree

There is no conflict between studies here, because every row comes from one method: the same search, run the same way for each phrase. The disagreement is between phrases. "Dictation" and "speech to text" are rising, "speech recognition" is flat, and some named products are falling. If you picked only one phrase you would reach a different conclusion. A reader who searched for "speech recognition" alone would call the topic stable. A reader who searched for "Wispr Flow" alone would call it exploding.

What the data cannot tell you

• It does not measure adoption. Hacker News skews toward developers and founders. It says what that audience discusses, not what writers, doctors, lawyers or students use.
• It does not measure sentiment. Nothing is filtered by meaning, so criticism counts the same as praise.
• It misses synonyms and misspellings. A product called by a nickname or typed another way will be undercounted.
• It reflects launches. A model release or a viral thread can push a phrase far above its normal level for a month, and one burst can account for much of a year's count.
• It cannot separate the senses of a word. Parakeet, moonshine and whisper all have other meanings.

What this means if you are choosing how to write

Do not read the counts as a ranking of quality. They show which tools developers are talking about, and that is a useful signal for one thing: where the ecosystem is moving. Right now it is moving toward local and open models, and toward apps that put them behind a hotkey. That is relevant if you want dictation that runs on your own machine and keeps audio off a server.

For a practical choice, look at the way you write, not the volume of talk. If you want a hotkey app for Mac or Windows, the products with the most recent discussion are the ones with active communities and fast fixes. If you only need occasional dictation, the built-in tools on your computer or phone may be enough. Our market overview puts these products in context with prices and features, and our guide to speech recognition explains how the engines underneath them differ. A rising count is a reason to try a tool, not a reason to trust it. Test two or three on your own voice and your own vocabulary before you commit.