← Speech to text (English): all libraries · Reader board
Vosk vs Whisper: which one makes fewer word errors?
Vosk 0.3.45 and Whisper 20250625 on the same 130 English clips with reference transcripts: audiobooks, read sentences, meetings (headset and table microphone) and earnings calls, 14.9 minutes from Speech test set (LibriSpeech, FLEURS, AMI meetings, Earnings-22 calls). Measured, not opinion; re-tested by itself on every new PyPI release. Updated 06 November 2026.
Whisper made fewer word errors: 17.6% vs 37.1%. Vosk broke (error, hang or crash) on 0 files that Whisper read; Whisper broke on 0 files that Vosk read.
| Library | Text from | 🔴 Fails where the other reads | 🟡 Empty where the other reads | ⏱ Hangs or crashes | ✅ Word error rate | Median ms/file | Total seconds |
|---|---|---|---|---|---|---|---|
| Vosk 0.3.45 | 129 | 0 | 1 | 0 | 37.1% | 918 | 138.2 |
| Whisper 20250625 | 130 | 0 | 0 | 0 | 17.6% | 1142 | 164.9 |
By kind of clip: word error rate
| Clips | Vosk | Whisper |
|---|---|---|
| ami-close-mic 20 clips | 44.3% | 22.7% |
| ami-far-field 15 clips | 79.3% | 49.0% |
| earnings-calls 25 clips | 60.4% | 26.6% |
| fleurs 20 clips | 32.5% | 12.9% |
| librispeech-clean 30 clips | 10.3% | 3.6% |
| librispeech-other 20 clips | 23.8% | 11.5% |
Files only Vosk reads (0)
None.
Files only Whisper reads (1)
Don't want to run it yourself? I also run Audio & Video Transcriber as a pay-per-use API on Apify (it uses faster-whisper). Disclosure: that tool is mine; the rankings above come only from the measurements.
What this does not tell you
This compares not breaking: whether each library gets any text out of each file. It does not rank text quality (reading order, tables, headings), and speed is one GitHub-hosted runner, default settings, no OCR switched on. If you need structure for RAG, also look at the output on your own files.
How this is measured
- Each library runs a short recipe from its own docs (board/targets.py):
open the file, get all text. Installed with the extras the docs recommend for reading
(
pypdf[full]: AES decryption, Brotli), nothing else. One worker process per CPU; per file 60 s / 120 s / 180 s / 300 s / 600 s at most (the any-document table allows longer: some tools load models), killed above the memory limit. - Fails where another reads: the library raised an error, hung or crashed on a file from which another library in the same table got text. Files that every library fails on (encrypted, truncated, fuzzed) are not counted against anyone.
- Empty where another reads: no error, but no text. This is often the recipe, not the library: for example
python-docx's paragraphs+tables recipe skips headers, text boxes and nested tables. Check the file list. - Any document → text: a file type the tool's docs don't list, or that the tool itself rejects as unsupported, counts as "not supported", not as a failure (lists in board/libs.json). Returning a file's raw RTF source counts as empty, for every tool. Each tool runs with default settings: no OCR switched on by hand.
- OCR: each image comes with the human-written ground-truth text published with its dataset. Words found is the share of its words that appear anywhere in the engine's output (reading order doesn't matter); an image that fails or comes back empty counts as 0. The ground truth of receipts and document pages covers only their annotated text, so a character error rate would punish engines for reading more than was annotated; it is in the raw results but not ranked. English models, CPU, default settings.
- Speech: each clip comes with its dataset's reference transcript. Both texts go through Whisper's English text normalizer (case, punctuation, "twenty" = "20"); word error rate is all substituted, missing and extra words over all reference words, the usual corpus-level WER. A clip that fails or comes back empty counts all its words as errors. The three Whisper implementations use the same base.en model; each other engine its documented default model. English, CPU, default decoding settings.
- Language detection: each text's language is known from Tatoeba. An answer counts as correct if the library's code matches after mapping variants to ISO 639-1 (zh-cn → zh, nb → no, iw → he). Close languages count as different: Croatian detected as Bosnian or Serbian, or Norwegian as Danish, is wrong. One call per text with default settings and all languages enabled.
- PDF → Markdown: pages from AI2's olmOCR-bench (ODC-BY), scored with its own checks (vendored, Apache-2.0): a sentence must be present, a page header or footer absent, two passages in the right order, a table cell next to the right neighbour. Checks passed is all passed checks over all checks; a page that fails or comes back empty passes none. Math pages are left out (their checks need a browser to render formulas). Default settings: whatever OCR a tool does on its own, it does.
- Text quality (order, tables, layout) is not ranked in the other tables; they are about not breaking. The raw results have every output's size and hash.
- Same files for every version: the test collections are pinned to one commit. GitHub-hosted runner, so times are comparable within one table, roughly.