Robust speech recognition can adequately detect speech under adverse conditions like noisy environments or in scratchy recordings. This can have important applications in a number of areas, such as law enforcement or the design of hearing aids, for example. Research and development into this topic occurs at academic institutions, private companies, and charitable organizations with an interest in this field all over the world. Careers in this field are open to people like sound engineers, computer programmers, and audiologists.
Conventional speech recognition suffers from the problem of having been designed for ideal environments. An algorithm can recognize speech if it occurs in a quiet environment with little to no background noise, and if the speaker clearly articulates the words. Such programs can struggle with accents that they haven't learned, and they also tend to break down in environments with lots of background noise. The world is often noisy, and thus such equipment can be of limited use in some settings without robust speech recognition.
In dictation, for example, most systems rely on a microphone worn close to the mouth, to allow the speaker's voice to dominate so the program can accurately process the speech. Speech recognition used in applications like remote listening for law enforcement, hearing aid design, and restoration of historic recordings can also have difficulty with background noise. Robust speech recognition involves the development of algorithms that can process and discard this noise to leave just the speech.
This requires complex computing abilities. Noisy environments can contain a wide variety of sounds, making it hard to simply create a pass filter that would cut out a range of noise. The filter might not catch all the problem noises, and could potentially interfere with the speech, as well. In robust speech recognition, programmers work to develop programs that can identify speech and separate it out from other tracks of sound. Once separated, it may be subjected to another pass to clean up the signal, allowing the program to run a normal speech recognition algorithm to determine what is being said.
Accurate speech recognition can be important for automated menus, dictation, and other real-time applications. The development of robust speech recognition can also help with the creation of hearing aids and software that and pinpoint human voices in a hum of other noise, and transmit just these to the listener. This makes speech recognition more useful in environments like crowded parties and events where multiple sounds may compete, potentially drowning out voices for listeners relying on speech recognition.
