Signal noise reduction using magnitude-domain spectral...

Details Signal noise reduction using magnitude-domain spectral... Signal noise reduction using magnitude-domain spectral...

2000-02-29

2004-10-12

Dorvil, Richemond (Department: 2654)

Data processing: speech signal processing, linguistics, language

Speech signal processing

For storage or transmission

C704S233000, C381S094300

Reexamination Certificate

active

06804640

ABSTRACT:

FIELD OF THE INVENTION
The present invention pertains to techniques for reducing noise in a signal. More particularly, the present invention relates to techniques for reducing noise in a signal representing speech.
BACKGROUND OF THE INVENTION
In a speech recognition system, the presence of noise in an input speech signal can degrade recognition accuracy. Noise can be introduced from many different sources and may be introduced through either acoustic coupling or electrical coupling. Acoustic coupling of noise into a speech signal might occur when a speaker is located in a noisy setting, such as a busy office environment. Electrical coupling of noise may result from electromagnetic radiation emitted by electrical devices in proximity to components of the speech recognition system. Various techniques are known for reducing noise in a speech signal. However, these techniques generally are not adaptable, or are not sufficiently adaptable, to the amount of noise in the speech signal at any given time. A typical consequence of this shortcoming is that a given noise reduction technique may perform adequately in a noisy environment but perform poorly in a low-noise environment. Such techniques, therefore, tend not to be very flexible in terms of handling signals under a variety of conditions. In addition, prior art noise reduction techniques may not be capable of operating upon individual frequency components of a signal. Furthermore, many noise reduction techniques used in speech recognition handle the beginning of a sentence very inefficiently, since few samples have been observed at that point in time, and a signal-to-noise ratio is difficult to estimate accurately at that time.
SUMMARY OF THE INVENTION
The present invention includes a method and apparatus for reducing noise in data representing an audio signal, such as a speech signal. For each of multiple frequency components of the audio signal, a spectral magnitude of the data and an estimate of noise in the data are computed. The estimate of noise is scaled by a noise scale factor that is a function of the corresponding frequency component, to produce a scaled noise estimate. The scaled noise estimate is subtracted from the spectral magnitude to produce cleaned audio data. The scale factor may be also be a function of an absolute noise level for each frequency component, a signal-to-noise ratio for each frequency component, or both.
The noise reduction technique operates well on both noisy input signals and clean input signals. The technique can be implemented as a real-time method of reducing noise in a speech signal, which begins the noise reduction process immediately when the speech signal becomes available, thereby reducing undesirable delay between production of the speech signal and its recognition by a speech recognizer.
Other features of the present invention will be apparent from the accompanying drawings and from the detailed description which follows.


REFERENCES:
patent: 4630304 (1986-12-01), Borth et al.
patent: 4811404 (1989-03-01), Vilmur et al.
patent: 4897878 (1990-01-01), Boll et al.
patent: 5604839 (1997-02-01), Acero et al.
patent: 5668927 (1997-09-01), Chan et al.
patent: 5684921 (1997-11-01), Bayya et al.
patent: 5692103 (1997-11-01), Lockwood et al.
patent: 5737407 (1998-04-01), Graumann
patent: 5742927 (1998-04-01), Crozier et al.
patent: 5806025 (1998-09-01), Vis et al.
patent: 5943429 (1999-08-01), Handel
patent: 5950154 (1999-09-01), Medaugh et al.
patent: 5963899 (1999-10-01), Bayya et al.
patent: 6098038 (2000-08-01), Hermansky et al.
patent: 6266633 (2001-07-01), Higgins et al.
patent: 6317709 (2001-11-01), Zack
patent: 6336090 (2002-01-01), Chou et al.
patent: 6349278 (2002-02-01), Krasny et al.
patent: 6351731 (2002-02-01), Anderson et al.
patent: 6477489 (2002-11-01), Lockwood et al.
patent: 6671667 (2003-12-01), Chandran et al.
patent: 0 534 837 (1993-03-01), None
patent: WO 99/14738 (1999-03-01), None
Lockwood et al., “Experiments with a Nonlinear Spectral Subtractor (NSS), Hidden Markov Models and the projection, for robust speech recognition in cars,” Jun. 1992, Speech Communication 11 (1992), Nos. 2-3, pp. 215-228.*
Kuo-Guan Wu, “Efficient speech enhancement using spectral subtraction for car hands-free applications,” International Conference on Consumer Electronics, 2001, Jun. 19-21, 2001, pp. 220 to 221.*
Steven F. Boll, “Suppression of Noise in Speech Using the Saber Method,” IEEE International Conference on Acoustics, Speech and Signal Processing, 1978, pp. 606-609.
M. Berouti, et al., “Enhancement of Speech Corrupted by Acoustic Noise,” Proceedings, IEEE Conference on Acoustics, Speech and Signal Processing, pp. 208-211, Apr. 1979.

Affiliated with

Also associated with

Rating

Signal noise reduction using magnitude-domain spectral... does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Signal noise reduction using magnitude-domain spectral..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Signal noise reduction using magnitude-domain spectral... will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-3276918