Speech-to-text translators (commonly abbreviated to "STT") are a class of speech recognition software, used to input and format dictated messages for a variety of applications.

Although STTs can be used in conjunction with language translation systems, the "translation" referred to is from spoken to written language. Depending on the language concerned, this may be as simple as automated dictation, or involve complex processes such as reformatting, interpolation, syntax adjustment, and so on in languages, or language-to-language configurations, where the written form is not an unaltered transcript of the spoken. STTs are also often capable of adding varying levels of "polish" to input, such as eliminating extraneous words or phrases, correcting grammar and syntax, and translating localised content such as time/date references into standardised forms.

Many STTs are built into omni-tool modules for use away from terminals; due to the limited storage and processing capacity, and the complexity of many languages, this can cause them to fall short of a perfect translation. Rather than manually proof-read the output, many users operating limited STTs in environments where informality is permissible - such as extranet forums - set their systems to notify readers of their operations (often inserting "<STT>" at the beginning of their output) so that any anomalies in the text are understood by the reader to be arfetacts of the STT system.

Some STT systems have a reputation for inconveniently loose interpretations of their auto-activate functions, resulting in users posting transcripts online without intending to.