15. How to create a clean Lombard speech database using loudspeakers
-
Christian Lüke
Abstract
The Lombard effect is a phenomenon of increasing the vocal effort that people develop due to, amongst others, the presence of background noise. For example, if the background noise level rises, both pitch and loudness of the voice usually increase, but also the duration of individual speech sounds and the diction may change. In order to investigate this effect in different noise scenarios within either a soundproof room or a car, a reproducible and easily changeable acoustic ambiance simulation has been developed and is presented in this chapter. The noise simulation uses loudspeakers instead of the usual approachwhich uses closed headphones. This increases the realism of the noise scenario for the subject and consequently also the ecological validity of the Lombard speech recordings. To remove the simulated noise from the recordings of the subject’s microphone signal, multichannel signal processing has been implemented to achieve nearly the same speech quality as with a common headset-based noise simulation.
Abstract
The Lombard effect is a phenomenon of increasing the vocal effort that people develop due to, amongst others, the presence of background noise. For example, if the background noise level rises, both pitch and loudness of the voice usually increase, but also the duration of individual speech sounds and the diction may change. In order to investigate this effect in different noise scenarios within either a soundproof room or a car, a reproducible and easily changeable acoustic ambiance simulation has been developed and is presented in this chapter. The noise simulation uses loudspeakers instead of the usual approachwhich uses closed headphones. This increases the realism of the noise scenario for the subject and consequently also the ecological validity of the Lombard speech recordings. To remove the simulated noise from the recordings of the subject’s microphone signal, multichannel signal processing has been implemented to achieve nearly the same speech quality as with a common headset-based noise simulation.
Chapters in this book
- Frontmatter I
- Contents V
- Preface XIII
- List of contributing authors XIX
-
Part I: Vehicle System and Safety
- 1. Analysis of in-vehicle speech activity towards driver safety assessment 3
- 2. Stochastic behavior modeling for driver assistance using stream data processing 19
- 3. Using real road driving data to calibrate a model of front-end collision risk 37
-
Part II: Driver Modeling
- 4. Driver mirror-checking action detection 55
- 5. Probabilistic driver modeling 77
- 6. Driving distance based analysis of driving maneuvers 99
- 7. Correlation of neurophysiological measurement of anxiety and driving behavior 111
- 8. Adaptation techniques for stochastic driver behavior modeling 123
- 9. Integrated modeling of driver gaze and vehicle operation behavior during lane changes 133
-
Part III: Signal Processing for HVI
- 10. Speaker activity detection for distributed microphone systems in cars 145
- 11. Speech enhancement employing feature domain reconstruction for robust in-vehicle speech recognition 161
- 12. Driver adaptive prediction for pedestrian detectability using in-vehicle camera images 171
- 13. An audio-visual in-car corpus “CENSREC-2-AV” for robust bimodal speech recognition 181
- 14. Array-based speech enhancement for microphones on seat belts 191
- 15. How to create a clean Lombard speech database using loudspeakers 209
- Index 227
Chapters in this book
- Frontmatter I
- Contents V
- Preface XIII
- List of contributing authors XIX
-
Part I: Vehicle System and Safety
- 1. Analysis of in-vehicle speech activity towards driver safety assessment 3
- 2. Stochastic behavior modeling for driver assistance using stream data processing 19
- 3. Using real road driving data to calibrate a model of front-end collision risk 37
-
Part II: Driver Modeling
- 4. Driver mirror-checking action detection 55
- 5. Probabilistic driver modeling 77
- 6. Driving distance based analysis of driving maneuvers 99
- 7. Correlation of neurophysiological measurement of anxiety and driving behavior 111
- 8. Adaptation techniques for stochastic driver behavior modeling 123
- 9. Integrated modeling of driver gaze and vehicle operation behavior during lane changes 133
-
Part III: Signal Processing for HVI
- 10. Speaker activity detection for distributed microphone systems in cars 145
- 11. Speech enhancement employing feature domain reconstruction for robust in-vehicle speech recognition 161
- 12. Driver adaptive prediction for pedestrian detectability using in-vehicle camera images 171
- 13. An audio-visual in-car corpus “CENSREC-2-AV” for robust bimodal speech recognition 181
- 14. Array-based speech enhancement for microphones on seat belts 191
- 15. How to create a clean Lombard speech database using loudspeakers 209
- Index 227
