A large share of Romanian text online is typed without diacritics, and readers cope, but the words are not the same: "fată" is a girl and "fata" is the girl, "peste" means over and "pește" is a fish, "tată" and "țață" are different relatives. A transcript meant for publication, subtitles or a court file needs the marks. The text produced here carries them from the start, so there is no pass to add them afterwards.
Two details matter to anyone who processes the text further. The letters ș and ț are written with a comma below, which is the correct Romanian form, though some older fonts and systems expect the cedilla variants ş and ţ and may display boxes. And the sound written â inside a word and î at its beginning or end follows the current academy rule: "român", "cuvânt", "în", "a coborî". Texts that follow the pre-1993 convention ("romîn", "sînt") are not produced.
Romanian attaches the definite article to the end of the noun, and in speech the final -l is almost never pronounced: people say "omu" and "băiatu" and write "omul" and "băiatul". The transcript restores the written form. The opposite case is the number of i at the end of a word, where "copii", "copiii" and "copi" sound alike; the model decides by grammar and occasionally writes one i too few or too many.
Hyphenated clitic groups are the classic Romanian spelling trap: "s-a" and "sa", "l-am", "mi-a", "într-o", "nu-i", "ne-am". They are handled from context and are usually correct. Fast, informal speech, where "pe care îl" collapses into "pe care-l" or "ce să fac" into something much shorter, is where a hyphen ends up in the wrong place. Numbers are written in digits; check amounts in lei and euro, dates and CNP-style identifiers against the audio.
Standard Romanian as heard in Bucharest and in national media is recognised best. Moldavian speech from Iași, with "ghini" for "bine" and "șî" for "și", is usually written in the standard form. Transylvanian speech, slower and with words such as "no" and "servus", comes through well. Strong Maramureș or Oltenian speech, with its simple perfect ("fusei", "văzui"), is transcribed readably with more corrections needed.
Recordings from the Republic of Moldova are transcribed as Romanian in Latin script. Russian words and whole Russian sentences dropped into the conversation are the difficulty there: short insertions are often written by sound in Romanian letters, and longer switches may be skipped or garbled. The same applies to Romanian speakers abroad who mix in Italian, Spanish or English. Office English ("deadline", "call", "feedback") is kept in English.
Romanian recordings uploaded here are mostly journalistic and research interviews, company and association meetings, online courses and webinars, sermons, podcasts and talk shows from YouTube. Manele, folk and pop lyrics sung over music transcribe poorly; this is a tool for speech. Guests get 10 transcriptions of up to 3 hours each and read a free preview; the free allowance is 5 hours in total, and a free account keeps the full transcripts. AI chat with the recording, automatic summaries and PDF or Word export are in the paid plans, and subtitles can be translated into other languages.
Audio is processed on our own servers with our own speech models and is not sent to a third-party speech API, which keeps the handling of interviews and internal meetings simple to explain under GDPR.