How much research is published on speech recognition each year?

The scholarly index OpenAlex lists 28,220 works published in 2025 that mention "speech recognition" ¹. That is the broadest count we track, and it covers papers, preprints, books and theses. The narrower phrase "automatic speech recognition" appears in 10,003 works from the same year ².

Growth is steep. The same index shows the phrase in 940 works for 1990 and in the 2025 figure above, across 36 yearly readings. Every topic we track except the smallest ones has grown since 2020, and the fastest growth is in the newest vocabulary.

Speech recognition research is now a large, steady stream

Two phrases carry most of the volume. "Speech recognition" leads, and "automatic speech recognition" is the next largest of the speech phrases we track. Both rose clearly between 2020 and 2025. The table lists each topic with its 2025 count and its count five years earlier.

Showing 8 of 10 rows. See every row in the dataset.

The next chart puts the yearly totals side by side.

Researchers reached for the broad phrase in 1990, and its use has risen almost every year since. The long view is below, from 1990 to 2025.

Speech-to-text and voice assistants grew fastest

Not every topic grew at the same pace. "Speech-to-text" appears in 9,278 works from 2025 ³, up from 2,643 in 2020. It was close to absent in 1990, with 37 works. "Voice assistant" appears in 6,728 works ⁴, up from 1,694 in 2020, and it had a single work in 1990.

Both phrases grew much faster over five years than the broad phrase "speech recognition," so the field is not only getting bigger, it is shifting toward products people use: assistants, transcription and text output.

Dictation research grows slowly, text entry moderately

"Dictation" appears in 3,139 works from 2025 ⁵, against 2,520 in 2020. That is the slowest growth among the larger topics, and it was already well established in 1990, with 171 works. Part of the reason is that many works use the word for language teaching or for clinical notes, not for software that types what you say.

"Text entry," the human-computer interaction term for keyboards, touch typing and voice input, appears in 1,508 works ⁶, up from 871 in 2020. "Voice typing" is much smaller, at 475 works ⁷, up from 233 in 2020. The third chart shows the smaller topics, including Whisper.

"Whisper speech" appears in 293 works from 2025 ⁸, up from 90 in 2020. The phrase also matches research on whispered speech, so it is not a clean count of work on the Whisper model.

Applied niches are tiny in this count

Phrases for specific uses return few works. "Medical speech recognition" appears in 20 works from 2025 ⁹, and "speech recognition accessibility" in 4 ¹⁰. These small numbers do not mean there is little research on those uses. Authors usually write about them in other words, such as "clinical documentation" or "assistive technology," which an exact phrase search does not catch.

What these numbers do and do not measure

Each figure is a count of works in OpenAlex whose text contains the quoted phrase, by publication year. Nothing is screened by hand. Keep four limits in mind:

• A phrase match is not relevance. A work counts if it uses the words anywhere the index looks, even when speech recognition is a side topic.
• Recent years are incomplete. OpenAlex keeps adding works, so the latest years grow for a while after they end. Each row carries the warning to compare topics and years, not exact totals.
• Phrases overlap. A work about automatic speech recognition also counts under speech recognition, so the rows must not be added together.
• Counts say nothing about quality. A rise in works does not show that tools got more accurate. For accuracy, see the guide to the history of speech recognition.

The dataset tracks 10 data points in all, updated monthly, so the figures on this page are always the latest reading. The full table, with each source, is in the full dataset.

What this means if you are choosing how to dictate

Research volume is a sign of a mature field, not a buying guide. The growth in speech-to-text and voice assistants matches what you see in products: transcription and voice input are now built into phones and computers, and the built-in tools are often enough for short messages and notes.

Choose by your own test. Dictate a few paragraphs in the place you actually write, in your accent and your noise level, and count the corrections. If you need text to stay on your own device, check where each tool processes audio. Yap, which we make, runs dictation on your device for Mac and Windows, but a free built-in tool is a fair starting point.

For a wider view of the subject, start with our overview of speech recognition, read the history of speech recognition, or open the full dataset to see every row.