How fast is Romanian spoken?
Romanian is the Romance language of Romania and Moldova, Italian's eastern relative. Every dot on the line below is one person reading sentences aloud: read Romanian runs at about five and a half syllables a second, the same pace as Italian measured the same way. At the end, any voice can be timed and placed on the same line.
265 voices on one line#
Volunteers for Mozilla Common Voice read printed sentences aloud, and other volunteers listened and validated the recordings. For every person with at least five kept recordings, one number was computed: the sentence's syllables divided by the speaking time, with pauses longer than 150 milliseconds taken out of the duration.
The speaker in the middle reads at 5.6 syllables per second; half of the speakers sit between 5.1 and 6.0. This is reading, not conversation, and the volunteers are not a sample of a country: the number describes the people measured, in the situation measured.
The close relative, measured with the same ruler#
Of the languages measured in the 2019 study, Italian is Romanian's close relative. In the same cold-reading recordings, 2,977 Italian speakers have their median at 5.3 syllables per second. Their cloud has almost the same shape and spread as the Romanian one.
Romanian sits slightly ahead in every sentence-length class; after the correction from the hand-counted syllables, the lead is about six per cent. For which related language is closest to Romanian by other measures, see “Which language is closest to Romanian?”.
Voices differ more than the two medians do#
The slowest kept Romanian speaker reads at 3.5 syllables per second, the fastest at 7.9: a ratio of 2.3 between people reading the same kind of sentences. Three of the voices can be heard here, each with its pace printed beside it.
Between the Romanian and Italian medians, measured the same way, the ratio is 1.05. In this corpus, the spread between people is far wider than the distance between the two medians.
Three recordings from the line above (Mozilla Common Voice, CC0), brought to the same loudness:
In another corpus, the number is different#
In both languages the pace rises with the length of the utterance: for sentences of 6–10 syllables the Romanian median is 4.9; for stretches of 31–60 syllables it reaches 6.4. One sentence read cold, at home, is the slowest rung measured here.
On the next rung, prepared speeches in the European Parliament (the VoxPopuli corpus) put Romanian at 6.5 syllables per second (49 speakers) and Italian at 6.4 (77 speakers): level. The two corpora hold different speakers, different material and different conditions, so the gap between rungs cannot be put down to the situation alone. A number without its situation says nothing, which is why every value on this page keeps its rung beside it.
On the 2019 study's scale, by a bridge#
The 2019 study by Coupé and colleagues (Science Advances) recorded ten speakers each for 17 languages, reading long texts after reading them several times first. Italian came out at 7.16 syllables per second. The same Italian, measured here cold, comes out at 5.3: 25.6% below the study's scale. The parliamentary rung brings it to 9.9% below that scale.
Romanian was not among the 17 languages. With Italian as the bridge — the Romanian-to-Italian ratio carried onto the study's scale, an assumption printed here rather than a measurement — Romanian would sit between 7.27 and 7.51 syllables per second: beside Italian, below Spanish (7.73), above French (6.88). The full chart of the 17 languages is on “Thirty-nine bits per second”.
| syllables/s | bits/s | |
|---|---|---|
| Japanese | 8.03 | 40.4 |
| Spanish | 7.73 | 42.0 |
| Basque | 7.54 | 36.4 |
| Finnish | 7.17 | 39.4 |
| Italian | 7.16 | 37.9 |
| Serbian | 7.15 | 39.1 |
| Korean | 7.12 | 39.6 |
| Catalan | 7.07 | 38.8 |
| Turkish | 7.05 | 37.6 |
| French | 6.88 | 45.9 |
| English | 6.34 | 44.9 |
| German | 6.09 | 37.0 |
| Hungarian | 5.87 | 34.6 |
| Mandarin Chinese | 5.86 | 40.8 |
| Cantonese | 5.57 | 36.4 |
| Vietnamese | 5.30 | 42.5 |
| Thai | 4.70 | 33.8 |
| Romanian ≈7.27–7.51 | 7.27–7.51 | 40.5–41.8 |
How much information one syllable carries#
Pace is half the story; the other half is how much information each syllable carries. With the 2019 estimator — the entropy of a syllable given the syllable before it in the word — computed on spelling syllables over Wikipedia text sampled identically (4,000,000 words per language), a Romanian syllable carries 6.11 bits, an Italian one 5.80 bits. On the same instrument the Italian control overshoots the published phonetic value by 0.51 bits; that gap stays printed beside every product. The difference between the two languages on this instrument, 0.31 bits, is smaller than the control's own offset: the within-method comparison stands, but its size should be read cautiously.
Multiplied by pace, at cold reading: Romanian 34.1 bits per second, Italian 30.9; on the parliamentary rung, 40.0 against 37.4. Across the same bridge as above, Romanian would sit on the study's scale between 40.5 and 41.8 bits per second, beside the 17-language mean of 39.15. Faster languages put less into each syllable; Romanian stands close to Italian on both factors, slightly denser per syllable.
Any voice, timed on the spot#
One sentence of 18 syllables, read aloud at a natural pace, in English. The sound stays on the device: the timing happens in the page, no recording is sent anywhere, and the result disappears when the page closes. The comparison line is 4,053 English speakers measured the same way (median 4.0 syllables per second).
In the morning, the train leaves the station and the fields slowly fill with light.
The comparison is with volunteers reading printed sentences, each at their own pace.
Age and sex, as far as they can be stated#
Of the 265 kept Romanian speakers, 189 declared an age or a sex. Women have their median at 5.6 syllables per second (36 speakers), men at 5.8 (151 speakers): the same direction as in the 2019 study, a small difference with almost fully overlapping distributions.
By declared age, speakers in their twenties read fastest (5.9, 121 speakers) and those in their forties more slowly (5.1, 21). Above fifty there are 5 speakers in total; no number printed here rests on them.
| syllables/s | n | |
|---|---|---|
| <20 | 5.7 | 11 |
| 20–29 | 5.9 | 121 |
| 30–39 | 5.7 | 30 |
| 40–49 | 5.1 | 21 |
| 50–59 | — | 4 |
| 60+ | — | 1 |
The hurried south and slow Moldavia: what these recordings cannot say#
A Romanian commonplace has the south — Bucharest, the capital, and the Wallachian plain — talking fast, and Moldavia, the north-eastern region reaching to Chișinău, the capital of the Republic of Moldova, talking slowly and melodically. The recordings used here cannot test the claim in either direction: the accent field is blank for 94% of the clips, and of the 16 filled-in labels, almost every one belongs to a single person.
An honest test of regional pace would need many speakers from each region reading the same sentences. Until such a corpus exists, the question stays open and no dot is placed on a map.
How it was measured, and how far it can be wrong#
The timing follows the 2019 definition: the standard pronunciation's syllables divided by the duration with pauses longer than 150 milliseconds removed. Syllables were counted from spelling by a stated rule in the project's code; because spelling does not settle every diphthong or hiatus, each sentence also carries a low–high band, and across the whole grid of reasonable choices on the same timing definition, the Romanian median stays between 5.4 and 6.5 syllables per second.
The measured errors, not assumed ones: a second, independent timing of 150 Romanian clips put the durations about 2.6% apart at the median; the automatic syllable count runs about 2% below a hand count on 100 Romanian sentences (and 1.2% on 30 Italian ones); Italian reproduced on this instrument lands 25.6% below the published value at the cold-reading rung and 9.9% below it on the parliamentary rung — a difference of situation, not of language.
Of the edition's 17,737 validated clips, the archive copy used held audio for 12,918; all 338 speakers remain represented. A median from five clips is still a rough estimate of a person's usual pace, and reading aloud is not conversation.
Sources and data#
The page's numbers are recomputed from the project's files at every publication. The sources, in order of use:
- Coupé, C., Oh, Y. M., Dediu, D., Pellegrino, F., “Different languages, similar encoding efficiency”, Science Advances 5 (2019). Full text; the 17 languages' values recomputed from the authors' published data file.
- Pellegrino, F., Coupé, C., Marsico, E., “A cross-language perspective on speech information rate”, Language 87 (2011). DOI.
- Mozilla Common Voice Corpus 17.0 (15 March 2024), Romanian, Italian and English (dev), CC0 licence. Dataset page; the corpus is described by Ardila et al., LREC 2020.
- VoxPopuli (European Parliament, 2009–2020), the Romanian and Italian test splits, CC0 licence. Corpus page.
- Wikipedia text (20231101 snapshot, CC BY-SA), used only to estimate information per syllable: wikimedia/wikipedia. English syllables from the CMU Pronouncing Dictionary 0.7b.
Recordings played on this page: Mozilla Common Voice, CC0; no speaker name or identifier is published.