Simultaneous Detection and

Estimation
Approach for Speech Enhancement

Presented by,
Libya Thomas,
M2 SP,
MBCET
1
 INTRODUCTION
• A simultaneous detection and estimation approach for speech
enhancement is proposed.
• A detector for speech presence in the short-time Fourier
transform domain is combined with an estimator.
• It jointly minimizes a cost function that takes into account both
detection and estimation errors.
• In addition a priori signal-to-noise ratio (SNR) estimator is
proposed for transient-noise environments.

2
 SIMULTANEOUS DETECTION &
ESTIMATION
• Speech enhancement systems often operate in the short-time Fourier
transform (STFT) domain.

• The spectral coefficients of the speech signal are generally sparse in

the STFT domain in the sense that speech is present only in some of
the frames.

• In our method a detector for the speech coefficients is combined

with an estimator, which jointly minimizes a cost function that takes
into account both estimation and detection errors.

3
 ASSUMPTIONS
• We consider two hypothesis-
H1 lk: Presence of speech
H0 lk : Absence of speech
• Noise is additive and uncorrelated.
• The objective of a speech enhancement system is to
reconstruct the spectral coefficients of the speech signal such
that under speech-presence a certain distortion measure
between the spectral coefficient and its estimate is minimized.

4
 SIMULTANEOUS DETECTION &
ESTIMATION cont.

• A decision space {ƞ0lk , ƞ1lk } is assumed for the detection

operation where under the decision ƞjlk, signal hypothesis Hj lk
is accepted and a corresponding estimate is
considered.

5
 SIMULTANEOUS DETECTION &
ESTIMATION cont.
• The detector is optimized with the knowledge of the specific
structure of the estimator, and the estimator is optimized in the
sense of minimizing a Bayesian risk associated with the
combined operations.

• Bayes risk of the two operations associated with simultaneous

detection and estimation is defined by

6
 A PRIORI SNR ESTIMATOR
• Used for estimation in the presence of transient noise.

• A lower bound for the a priori SNR which is necessary to

reduce noise is calculated.

• A priori SNR is defined under the assumption that H1 lk is true.

7
 A PRIORI SNR ESTIMATOR cont.
• Since the a priori SNR is highly dependent on the noise
variance, we first estimate the speech spectral variance.

• The a priori SNR is evaluated by taking the ratio of the speech

spectrum variance and the noise variance so that a lower
bound is obtained.

• The lower bound a priori SNR attenuates high energy transient

noise.

8
 CONCLUSION
• The proposed method uses a combined structure for detection and
estimation in STFT domain.
• The cost function is considered which combines both the detection
and estimation errors.
• Combined Bayes risk is minimized.
• In addition a priori estimator is proposed which is applicable in the
presence of transient noise.
• The algorithm ensure great noise reduction and improved speech
quality.

9
 REFERENCES
[1] A. Abramson and I. Cohen, "Simultaneous Detection and
Estimation Approach for Speech Enhancement," in IEEE
Transactions on Audio, Speech, and Language Processing, vol.
15,no.8,pp.2348-2359,Nov. 2007.doi: 10.1109/TASL.2007.904231
[2] S. F. Boll, “Suppression of acousting noise in speech using
spectral subtraction,” IEEE Trans. Acoust., Speech, Signal
Process., vol.ASSP-27, no. 2, pp. 113–120, Apr. 1979.
[3] M. Berouti, R. Schwartz, and J. Makhoul, “Enhancement of
speech corrupted by acoustic noise,” in Proc. IEEE Int. Conf.
Acoust., Speech Signal Process., ICASSP’79, Apr. 1979, vol. 4, pp.
208–211.

10