Speech or Voice Recognition involves the identification of discrete words in natural language from an analog or digital audio signal. The output of a speech recognition application is generally a text file, although systems may be embedded and packaged with NLU and or NLG so the intermediate text is never seen as a discrete output.
Early speech recognition systems were often limited to a domain specific vocabulary (e.g. medical transcription), and hampered by difficulty separating the voice signal from ambient noise. Today, the underlying algorithms are mature enough to perform well in home and office environments with modestly priced directional microphones.
Representative Vendors: Google, Gridspace, IBM, Microsoft Maluuba, Cisco (MindMeld), Nuance, Pop Up Archive, Rev.ai, SayIt, and Viv Labs.