A speech-to-text conversion module uses a central database of user speech profiles to convert speech to text. Incoming audio information is fragmented into numerous audio fragments based upon detecting silence. The audio information is also converted to numerous text files by any number of speech engines. Each text file is then fragmented into numerous text fragments based upon the boundaries established during the audio fragmentation. Each set of text fragments from the different speech engines corresponding to a single audio fragments is then compared. The best approximation of the audio fragment is produced from the set of text fragments; a hybrid may be produced. If no agreement is reached, the audio fragment and set the text fragments are sent to human agents who verify and edit to produce a final edited text fragment that best corresponds to the audio fragment. Fragmentation that produces overlapping audio fragments requires splicing of the final text fragments to produce the output text file.