A flexible and inexpensive real-time speech recognition system is described. It operates in the following modes: recognition of isolated words from a large vocabulary, and orthographic transcription of (eventually continuous) speech. The main parts are the acoustic processor module that transcribes speech into phonemes, a large-vocabulary lexical-access module that recognizes isolated words on the basis of these transcriptions, and a character-string processor module that produces orthographically edited text for Finnish and romanized Japanese from the erroneus transcriptions within unlimited vocabulary.