Noise Adaptive Training for Robust Automatic Speech Recognition

Editor's Notes

• #6: SpeakQ only works for text – not use to control entire computer
• #7: Visual Motor - The mechanical aspects of keyboarding are reduced significantly. Apart from the reduced motor load with reduced need to use the keyboard, there is also a reduced visual motor load since there is far less movement of the eyes from the keyboard to the screen. Spelling - The student who has good spelling and good site word recognition, has the potential of a high level of independent use of the program. However, for those students to have very poor spelling and who are unable to visually isolate incorrect words on the screen, additional add-on programs such as Keystone ScreenSpeaker are invaluable in allowing the students to use speech recognition. No longer do students need to have high levels of literacy in order to use NaturallySpeaking. As long as the student has the potential to use both NaturallySpeaking and Keystone ScreenSpeaker in conjunction with one another, there is potential for using speech recognition. Ergonomics - Using speech recognition reduces the amount of keyboard and mouse entry required. This allows the student to be more flexible in terms of positioning and posture and also allows for more movement or alteration of posture while working. In addition, the prevalence of repetitive stress injuries with extended use of the keyboard or mouse is significantly reduced. Students who have difficulty maintaining attention and posture, will find the more relaxed postural requirements of using speech recognition more conducive to working and producing written documents. Ergonomics and body posture including head, body, arms and legs are still very important while working. Students should still maintain a good body posture with appropriate height of chair, table and monitor. Hands-free use - While using speech recognition, the user has the option of dictating most or all of the text using the microphone as well as performing some or all computer commands and control functions. What this means is that the option of moving the mouse, opening and closing programs, moving from one place to the next on the computer and clicking on buttons are all options provided by speech input. Users with learning disabilities often experience subtle problems with fine motor coordination and control. These users may choose to perform more functions by voice rather than using the mouse or keyboard. Users with significant physical disabilities may want to use full command and control options available in NaturallySpeaking. Apart from the options available in NaturallySpeaking, there are other programs such as QPointer Voice which primarily offer command and control functions with speech input. For students with physical disabilities wanting to use voice input to control the mouse as well as for computer functions, the use of Dragon NaturallySpeaking Professional is suggested. NaturallySpeaking Professional allows for custom macros and commands to be added which make the program extremely powerful for users with significant physical disabilities. Users are able to add their own macros, or can save a lot of time and efforts by purchasing third-party add on commands modules such as the KnowBrainer Command Software by Lunis Orcutt. Detailed information regarding computer commands and controlled by voice is provided in the curriculum on this web site. Endurance - For those students who experience greater levels of fatigue while working, using speech recognition is of benefit since reduced physical energy expenditure is required with more flexible and relaxed postural requirements. The option of setting in a more comfortable seated position with greater body support, leaning backwards, results in increased endurance for written work and greater focus on the content of work rather than focus on controlling and maintaining posture and arm movements required for keyboarding or writing. Concentration and Attention - Learning how to use speech recognition and using the program over time requires good amounts of concentration and focus. Students who have concentration difficulties may find initial training difficult, and the need for sustained concentration for editing and corrections challenging. However, through work over many years with learning disabled students, it is noted that for many students the positive outcomes in terms of increased quality of work and rate of work, as well as the process of improved rate and production of text results in improved motivation and also improved focus of attention. Speech recognition can be a significant motivator for a student who is experienced years of failure with producing written work. In addition, although speech recognition requires the cognitive skills of memory for commands and understanding when and how to make corrections, it reduces the amount of divided attention normally found in regular keyboarding tasks. Regular keyboarding requires the user to focus attention not only on the screen and the text which appears, but also attention on the keyboard and location of individual keys, as well as focus on maintaining good body posture and alignment for keyboarding. Speech recognition actually reduces the amount of divided attention by allowing the user to focus on their speech and producing text / brainstorming with editing and corrections taking place at a later stage in the writing process. Reading and Speech - the use of speech output together with speech input (speech recognition) is an essential part of the whole speech recognition process. Speech output in Dragon NaturallySpeaking occurs in two different forms: Digitized speech - in which the user's voice is recorded and can be played back. As the user dictates his voice is recorded and during playback each word is indicated to show what has been produced as text related to the user's speech. Synthesized speech - this is purely a text-to-speech process in which the computer reads the text which has been highlighted on the screen. This is played back in a computer/synthesized voice. The value of digitized speech output is that each user, regardless of their consistency of articulation or speech, is able to have their own speech or dictation read back to them in their own voice as they had initially dictated it. This allows the user to determine if an error was an error of dictation or an error of recognition. Many times, people may make errors of articulation and not be aware of it until they play back their own dictation. Synthesized speech or screen reading, allows users to have difficulty reading or editing their work, to have the computer or read back the exact words on the screen. If the computer has made an error in recognizing words that was said, the user will realize these errors either through reading the words or three hearing the words read aloud through the synthesized speech option. Synthesized speech is extremely valuable for students who have difficulty reading or isolating text errors on the screen. NaturallySpeaking has synthesized speech reading built in to the program (preferred version and higher), but this will only read text that has been recognized on the screen. An additional programs such as Keystone ScreenSpeaker is required to read other parts of text in NaturallySpeaking such as the training texts and the correction lists. Pronunciation and Articulation - in previous versions of NaturallySpeaking, users with poor articulation and was inconsistent pronunciation found the use of speech recognition extremely frustrating and in most cases nonfunctional. With the improvements in the accuracy of the program, more and more individuals with speech and articulation problems are managing to use the program and use it more successfully. An individual with speech challenges will find facts additional training and higher levels of ongoing corrections are required. However, these may be acceptable trade-offs for someone who is unable to type fast or who has difficulty with their spelling. The potential for using speech recognition if a user has speech challenges can only be determined on a one-to-one basis with each individual. At trial of speech recognition with significant support and a structured training program as well as much practice may result in an acceptable, functional level of use. For users with more significant speech in payments, Dragon NaturallySpeaking can be used in spell mode, together with add-on word prediction software, so that NaturallySpeaking is almost being used as a keyboard with individual keystrokes provided by voice input rather than finger on a keyboard. Visual Motor - The mechanical aspects of keyboarding are reduced significantly. Apart from the reduced motor load with reduced need to use the keyboard, there is also a reduced visual motor load since there is far less movement of the eyes from the keyboard to the screen. Spelling - The student who has good spelling and good site word recognition, has the potential of a high level of independent use of the program. However, for those students to have very poor spelling and who are unable to visually isolate incorrect words on the screen, additional add-on programs such as Keystone ScreenSpeaker are invaluable in allowing the students to use speech recognition. No longer do students need to have high levels of literacy in order to use NaturallySpeaking. As long as the student has the potential to use both NaturallySpeaking and Keystone ScreenSpeaker in conjunction with one another, there is potential for using speech recognition. Ergonomics - Using speech recognition reduces the amount of keyboard and mouse entry required. This allows the student to be more flexible in terms of positioning and posture and also allows for more movement or alteration of posture while working. In addition, the prevalence of repetitive stress injuries with extended use of the keyboard or mouse is significantly reduced. Students who have difficulty maintaining attention and posture, will find the more relaxed postural requirements of using speech recognition more conducive to working and producing written documents. Ergonomics and body posture including head, body, arms and legs are still very important while working. Students should still maintain a good body posture with appropriate height of chair, table and monitor. Hands-free use - While using speech recognition, the user has the option of dictating most or all of the text using the microphone as well as performing some or all computer commands and control functions. What this means is that the option of moving the mouse, opening and closing programs, moving from one place to the next on the computer and clicking on buttons are all options provided by speech input. Users with learning disabilities often experience subtle problems with fine motor coordination and control. These users may choose to perform more functions by voice rather than using the mouse or keyboard.  Users with significant physical disabilities may want to use full command and control options available in NaturallySpeaking. Apart from the options available in NaturallySpeaking, there are other programs such as QPointer Voice which primarily offer command and control functions with speech input. For students with physical disabilities wanting to use voice input to control the mouse as well as for computer functions, the use of Dragon NaturallySpeaking Professional is suggested. NaturallySpeaking Professional allows for custom macros and commands to be added which make the program extremely powerful for users with significant physical disabilities. Users are able to add their own macros, or can save a lot of time and efforts by purchasing third-party add on commands modules such as the KnowBrainer Command Software by Lunis Orcutt.  Detailed information regarding computer commands and controlled by voice is provided in the curriculum on this web site.  Endurance - For those students who experience greater levels of fatigue while working, using speech recognition is of benefit since reduced physical energy expenditure is required with more flexible and relaxed postural requirements. The option of setting in a more comfortable seated position with greater body support, leaning backwards, results in increased endurance for written work and greater focus on the content of work rather than focus on controlling and maintaining posture and arm movements required for keyboarding or writing. Concentration and Attention - Learning how to use speech recognition and using the program over time requires good amounts of concentration and focus. Students who have concentration difficulties may find initial training difficult, and the need for sustained concentration for editing and corrections challenging. However, through work over many years with learning disabled students, it is noted that for many students the positive outcomes in terms of increased quality of work and rate of work, as well as the process of improved rate and production of text results in improved motivation and also improved focus of attention. Speech recognition can be a significant motivator for a student who is experienced years of failure with producing written work. In addition, although speech recognition requires the cognitive skills of memory for commands and understanding when and how to make corrections, it reduces the amount of divided attention normally found in regular keyboarding tasks. Regular keyboarding requires the user to focus attention not only on the screen and the text which appears, but also attention on the keyboard and location of individual keys, as well as focus on maintaining good body posture and alignment for keyboarding. Speech recognition actually reduces the amount of divided attention by allowing the user to focus on their speech and producing text / brainstorming with editing and corrections taking place at a later stage in the writing process. Reading and Speech - the use of speech output together with speech input (speech recognition) is an essential part of the whole speech recognition process. Speech output in Dragon NaturallySpeaking occurs in two different forms: Digitized speech - in which the user's voice is recorded and can be played back. As the user dictates his voice is recorded and during playback each word is indicated to show what has been produced as text related to the user's speech. Synthesized speech - this is purely a text-to-speech process in which the computer reads the text which has been highlighted on the screen. This is played back in a computer/synthesized voice. The value of digitized speech output is that each user, regardless of their consistency of articulation or speech, is able to have their own speech or dictation read back to them in their own voice as they had initially dictated it. This allows the user to determine if an error was an error of dictation or an error of recognition. Many times, people may make errors of articulation and not be aware of it until they play back their own dictation. Synthesized speech or screen reading, allows users to have difficulty reading or editing their work, to have the computer or read back the exact words on the screen. If the computer has made an error in recognizing words that was said, the user will realize these errors either through reading the words or three hearing the words read aloud through the synthesized speech option. Synthesized speech is extremely valuable for students who have difficulty reading or isolating text errors on the screen. NaturallySpeaking has synthesized speech reading built in to the program (preferred version and higher), but this will only read text that has been recognized on the screen. An additional programs such as Keystone ScreenSpeaker is required to read other parts of text in NaturallySpeaking such as the training texts and the correction lists. Pronunciation and Articulation - in previous versions of NaturallySpeaking, users with poor articulation and was inconsistent pronunciation found the use of speech recognition extremely frustrating and in most cases nonfunctional. With the improvements in the accuracy of the program, more and more individuals with speech and articulation problems are managing to use the program and use it more successfully. An individual with speech challenges will find facts additional training and higher levels of ongoing corrections are required. However, these may be acceptable trade-offs for someone who is unable to type fast or who has difficulty with their spelling. The potential for using speech recognition if a user has speech challenges can only be determined on a one-to-one basis with each individual. At trial of speech recognition with significant support and a structured training program as well as much practice may result in an acceptable, functional level of use. For users with more significant speech in payments, Dragon NaturallySpeaking can be used in spell mode, together with add-on word prediction software, so that NaturallySpeaking is almost being used as a keyboard with individual keystrokes provided by voice input rather than finger on a keyboard. Visual Motor - The mechanical aspects of keyboarding are reduced significantly. Apart from the reduced motor load with reduced need to use the keyboard, there is also a reduced visual motor load since there is far less movement of the eyes from the keyboard to the screen. Spelling - The student who has good spelling and good site word recognition, has the potential of a high level of independent use of the program. However, for those students to have very poor spelling and who are unable to visually isolate incorrect words on the screen, additional add-on programs such as Keystone ScreenSpeaker are invaluable in allowing the students to use speech recognition. No longer do students need to have high levels of literacy in order to use NaturallySpeaking. As long as the student has the potential to use both NaturallySpeaking and Keystone ScreenSpeaker in conjunction with one another, there is potential for using speech recognition. Ergonomics - Using speech recognition reduces the amount of keyboard and mouse entry required. This allows the student to be more flexible in terms of positioning and posture and also allows for more movement or alteration of posture while working. In addition, the prevalence of repetitive stress injuries with extended use of the keyboard or mouse is significantly reduced. Students who have difficulty maintaining attention and posture, will find the more relaxed postural requirements of using speech recognition more conducive to working and producing written documents. Ergonomics and body posture including head, body, arms and legs are still very important while working. Students should still maintain a good body posture with appropriate height of chair, table and monitor. Hands-free use - While using speech recognition, the user has the option of dictating most or all of the text using the microphone as well as performing some or all computer commands and control functions. What this means is that the option of moving the mouse, opening and closing programs, moving from one place to the next on the computer and clicking on buttons are all options provided by speech input. Users with learning disabilities often experience subtle problems with fine motor coordination and control. These users may choose to perform more functions by voice rather than using the mouse or keyboard.  Users with significant physical disabilities may want to use full command and control options available in NaturallySpeaking. Apart from the options available in NaturallySpeaking, there are other programs such as QPointer Voice which primarily offer command and control functions with speech input. For students with physical disabilities wanting to use voice input to control the mouse as well as for computer functions, the use of Dragon NaturallySpeaking Professional is suggested. NaturallySpeaking Professional allows for custom macros and commands to be added which make the program extremely powerful for users with significant physical disabilities. Users are able to add their own macros, or can save a lot of time and efforts by purchasing third-party add on commands modules such as the KnowBrainer Command Software by Lunis Orcutt.  Detailed information regarding computer commands and controlled by voice is provided in the curriculum on this web site.  Endurance - For those students who experience greater levels of fatigue while working, using speech recognition is of benefit since reduced physical energy expenditure is required with more flexible and relaxed postural requirements. The option of setting in a more comfortable seated position with greater body support, leaning backwards, results in increased endurance for written work and greater focus on the content of work rather than focus on controlling and maintaining posture and arm movements required for keyboarding or writing. Concentration and Attention - Learning how to use speech recognition and using the program over time requires good amounts of concentration and focus. Students who have concentration difficulties may find initial training difficult, and the need for sustained concentration for editing and corrections challenging. However, through work over many years with learning disabled students, it is noted that for many students the positive outcomes in terms of increased quality of work and rate of work, as well as the process of improved rate and production of text results in improved motivation and also improved focus of attention. Speech recognition can be a significant motivator for a student who is experienced years of failure with producing written work. In addition, although speech recognition requires the cognitive skills of memory for commands and understanding when and how to make corrections, it reduces the amount of divided attention normally found in regular keyboarding tasks. Regular keyboarding requires the user to focus attention not only on the screen and the text which appears, but also attention on the keyboard and location of individual keys, as well as focus on maintaining good body posture and alignment for keyboarding. Speech recognition actually reduces the amount of divided attention by allowing the user to focus on their speech and producing text / brainstorming with editing and corrections taking place at a later stage in the writing process. Reading and Speech - the use of speech output together with speech input (speech recognition) is an essential part of the whole speech recognition process. Speech output in Dragon NaturallySpeaking occurs in two different forms: Digitized speech - in which the user's voice is recorded and can be played back. As the user dictates his voice is recorded and during playback each word is indicated to show what has been produced as text related to the user's speech. Synthesized speech - this is purely a text-to-speech process in which the computer reads the text which has been highlighted on the screen. This is played back in a computer/synthesized voice. The value of digitized speech output is that each user, regardless of their consistency of articulation or speech, is able to have their own speech or dictation read back to them in their own voice as they had initially dictated it. This allows the user to determine if an error was an error of dictation or an error of recognition. Many times, people may make errors of articulation and not be aware of it until they play back their own dictation. Synthesized speech or screen reading, allows users to have difficulty reading or editing their work, to have the computer or read back the exact words on the screen. If the computer has made an error in recognizing words that was said, the user will realize these errors either through reading the words or three hearing the words read aloud through the synthesized speech option. Synthesized speech is extremely valuable for students who have difficulty reading or isolating text errors on the screen. NaturallySpeaking has synthesized speech reading built in to the program (preferred version and higher), but this will only read text that has been recognized on the screen. An additional programs such as Keystone ScreenSpeaker is required to read other parts of text in NaturallySpeaking such as the training texts and the correction lists. Pronunciation and Articulation - in previous versions of NaturallySpeaking, users with poor articulation and was inconsistent pronunciation found the use of speech recognition extremely frustrating and in most cases nonfunctional. With the improvements in the accuracy of the program, more and more individuals with speech and articulation problems are managing to use the program and use it more successfully. An individual with speech challenges will find facts additional training and higher levels of ongoing corrections are required. However, these may be acceptable trade-offs for someone who is unable to type fast or who has difficulty with their spelling. The potential for using speech recognition if a user has speech challenges can only be determined on a one-to-one basis with each individual. At trial of speech recognition with significant support and a structured training program as well as much practice may result in an acceptable, functional level of use. For users with more significant speech in payments, Dragon NaturallySpeaking can be used in spell mode, together with add-on word prediction software, so that NaturallySpeaking is almost being used as a keyboard with individual keystrokes provided by voice input rather than finger on a keyboard.
• #8: Users who exhibit borderline or questionable memory skills and concentration may be able to cope with the program given a modified or reduced commands set combined with assistance and reduced expectations in terms of the complexity of voice commands used. As you will note in the speech-recognition curriculum provided on customtyping.com, the commands used in stage three are a basic set of commands that all speech-recognition users should learn and master. Given the accurate use of these basic Stage Three commands, users can produce basic written documents.
• #9: As speech recognition programs have improved over the years, more and more people with articulation and pronunciation problems have experienced success in using the program. Current versions of the program seem to cope with a wider variety of speech patterns than previous versions. However a constant requirement over the years has been the need for consistency in speech patterns. For those users with more unusual speech patterns and articulation, additional training of the program may be required. However as the voice file is built and developed, the user's own unique, individual speech patterns are learned. The most important aspect here is that the words and phrases are said in the same Although many users on ventilators/respirators are extremely successful using speech recognition, they have learnt to control breathing and speaking so that their speech is consistent and breath sounds are controlled.
• #20: The Dragon Remote Microphone app turns the iPhone into a wireless microphone making it easier and more comfortable for individuals to use their Dragon Desktop software. Instead of using the microphone that comes in the retail box, users can now opt to use their iPhone as a microphone instead. This app can be used with Dragon NaturallySpeaking for the PC (v11.5 and higher) and Dragon Dictate for the Mac (V2.5 and higher). The Dragon Remote Microphone app turns the iPhone into a wireless microphone making it easier and more comfortable for individuals to use their Dragon Desktop software. Instead of using the microphone that comes in the retail box, users can now opt to use their iPhone as a microphone instead. This app can be used with Dragon NaturallySpeaking for the PC (v11.5 and higher) and Dragon Dictate for the Mac (V2.5 and higher). The Dragon Remote Microphone app turns the iPhone into a wireless microphone making it easier and more comfortable for individuals to use their Dragon Desktop software. Instead of using the microphone that comes in the retail box, users can now opt to use their iPhone as a microphone instead. This app can be used with Dragon NaturallySpeaking for the PC (v11.5 and higher) and Dragon Dictate for the Mac (V2.5 and higher). The Dragon Remote Microphone app turns the iPhone into a wireless microphone making it easier and more comfortable for individuals to use their Dragon Desktop software. Instead of using the microphone that comes in the retail box, users can now opt to use their iPhone as a microphone instead. This app can be used with Dragon NaturallySpeaking for the PC (v11.5 and higher) and Dragon Dictate for the Mac (V2.5 and higher). The Dragon Remote Microphone app turns the iPhone into a wireless microphone making it easier and more comfortable for individuals to use their Dragon Desktop software. Instead of using the microphone that comes in the retail box, users can now opt to use their iPhone as a microphone instead. This app can be used with Dragon NaturallySpeaking for the PC (v11.5 and higher) and Dragon Dictate for the Mac (V2.5 and higher). The Dragon Remote Microphone app turns the iPhone into a wireless microphone making it easier and more comfortable for individuals to use their Dragon Desktop software. Instead of using the microphone that comes in the retail box, users can now opt to use their iPhone as a microphone instead. This app can be used with Dragon NaturallySpeaking for the PC (v11.5 and higher) and Dragon Dictate for the Mac (V2.5 and higher).
• #34: Over time, training times for initial training have decreased and the program itself has become more accurate.