Search

Effective noise reduction and speech enhancement algorithms have great potential to enhance lives of hearing aid users by restoring speech intelligibility. An open problem in today’s commercial hearing aids is how to take into account users’ preferences, indicating which acoustic sources should be suppressed or enhanced, since they are not only user-specific but also depend on many situational factors. In this paper, we develop a fully probabilistic approach to “situated soundscaping”, which aims at enabling users to make on-the-spot (“situated”) decisions about the enhancement or suppression of individual acoustic sources. The approach rests on a compact generative probabilistic model for acoustic signals. In this framework, all signal processing tasks (source modeling, source separation and soundscaping) are framed as automatable probabilistic inference tasks. These tasks can be efficiently executed using message passing-based inference on factor graphs. Since all signal processing tasks are automatable, the approach supports fast future model design cycles in an effort to reach commercializable performance levels. The presented results show promising performance in terms of SNR, PESQ and STOI improvements in a situated setting.

Objective: Preterm birth is a large-scale clinical problem involving over 10% of infants. Diagnostic means for timely risk assessment are lacking and the underlying physiological mechanisms unclear. To improve the evaluation of pregnancy before term, we introduce dedicated entropy measures derived from a single-channel electrohysterogram (EHG). Methods: The estimation of approximate entropy (ApEn) and sample entropy (SampEn) is adjusted to monitor variations in the regularity of single-channel EHG recordings, reflecting myoelectrical changes due to pregnancy progression. In particular, modifications in the tolerance metrics are introduced for improving robustness to EHG amplitude fluctuations. An extensive database of 58 EHG recordings with 4 monopolar channels in women presenting with preterm contractions was manually annotated and used for validation. The methods were tested for their ability to recognize the onset of labor and the risk of preterm birth. Comparison with the best single-channel methods according to the literature was performed. Results: The reference methods were outperformed. SampEn and ApEn produced the best prediction of delivery, although only one channel showed a significant difference (p

Fuzzy commitment and syndrome-based schemes are two well-known helper data schemes used to bind and generate, respectively, a secret key to/from SRAM-PUF observations. To allow the decoder to reconstruct this secret key from a new (verification) observation of an SRAM-PUF, an encoder has to generate so-called helper data. This helper data is a function of an SRAM-PUF enrollment observation and, in case of fuzzy commitment, the secret key. The helper data is assumed to be public and thus must leak no information about the secret key. It is known that both schemes can achieve secrecy capacity equal to the mutual information between enrollment and verification SRAM-PUF observations at zero secrecy leakage, when the observations are unbiased and a single enrollment is performed. We study here the situation when multiple SRAM-PUF observations are used to create multiple secret keys. First, we introduce a symmetry property for multiple SRAM-PUF observations. For such symmetric SRAM-PUFs, we show that, in both helper data schemes, the helper data corresponding to multiple SRAM-PUF observations provide no information about any of the secret keys.

In biometric privacy-preserving authentication systems that are based on key-binding, two terminals observe two correlated biometric sequences. The first terminal selects a secret key, which is independent of the biometric data, binds this secret key to the observed biometric sequence and communicates it to the second terminal by sending a public message. This message should only contain a negligible amount of information about the secret key, but also leak as little as possible about the biometric data. Current approaches to realize such biometric systems use fuzzy commitment with codes that, given a secret-key rate, can only achieve the corresponding privacy-leakage rate equal to one minus this secret-key rate. However, the results in Willems and Ignatenko [2009] indicate that lower privacy leakage can be achieved if vector quantization is used at the encoder. In this paper we study the use of convolutional and turbo codes applied in fuzzy commitment and its modifications that realize this.

We present models for authentication in biometric settings. We could determine the trade-off between false-acceptance exponents and the so-called privacy-leakage rate. In this way we extend the Ahlswede-Csiszar secret-key generation result [1993] and more specifically the secret-key rate vs. privacy-leakage rate trade-off, studied by Ignatenko and Willems [2009].

Fuzzy commitment of Juels and Wattenberg 1999 is a popular technique for designing secure systems based on noisy data. The scheme is easy to implement using standard error-correcting codes. However, secrecy of this scheme is only guaranteed when input data are generated by uniform i.i.d. sources, while typical input data (PUFs and biometrics) are not uniform. In this paper we address the problem of extracting robust independent uniformly distributed bits out of noisy data that can be used as entries to fuzzy commitment. The proposed techniques can serve as a building block of secure fuzzy commitment systems. © 2009 IEEE.

The fundamental secret-key rate vs. privacy-leakage rate trade-offs for secret-key generation and transmission for i.i.d. Gaussian biometric sources are determined. These results are the Gaussian equivalents of the results that were obtained for the discrete case by the authors and independently by Lai et al. in 2008. Also the effect that binary quantization of the biometric sequences has on the ratio of the secret-key rate and privacy-leakage rate is considered. It is shown that the squared correlation coefficient must be increased by a factor of pi2/4 to compensate for such a quantization action, for values of the privacy-leakage rate that approach zero, when the correlation coefficient is close to zero.

Motivated by Maurer [1993], Ahlswede and Csiszar [1993] introduced the concept of secret sharing. In their source model two terminals observe two correlated sequences. It is the objective of both terminals to form a common secret by interchanging a public message (helper data), that should contain only a negligible amount of information about the secret. Ahlswede and Csiszar showed that the maximum secret key rate that can be achieved in this way is equal to the mutual information between the two source outputs. In a biometric setting, where the sequences correspond to the enrollment and authentication data, it is crucial that the public message leaks as little information as possible about the biometric data, since compromised biometric data cannot be replaced. We investigate the fundamental trade-offs for four biometric settings. The first one is the standard (Ahlswede-Csiszar) secret generation setting, for which we determine the secret key rate - privacy leakage region. Here leakage corresponds to the mutual information between helper data and biometric enrollment sequence conditional on the secret. In the second setting the secret is not generated by the terminals but independently chosen, and transmitted using a public message. Again we determine the region of achievable rate - leakage pairs. In setting three and four we consider zero-leakage, i.e. the public message contains only a negligible amount of information about the secret and the biometric enrollment sequence. To achieve this a private key is needed which can be observed only by the terminals. We consider again both secret generation and secret transmission and determine for both cases the region of achievable secret key rate - private key rate pairs.

We propose methods to estimate the secrecy-rate of fuzzy sources (e.g. biometrics and Physical Unclonable Functions (PUFs)) using context-tree weighting (CTW, Willems et al. [1995]). In this paper we focus on PUFs. In order to show that our estimates are realistic we first generalize Maurer's [1993] result to the ergodic case. Then we focus on the fact that the entropy of a stationary two-dimensional structure is a limit of a series of conditional entropies, a result by Anastassiou and Sakrison [1982]. We extend this result to the conditional entropy of one two-dimensional structure given another one. Finally we show that the general CTW-method approaches the source entropy also in the two-dimensional stationary case. We further extend this result to the two-dimensional conditional entropy. Based on the obtained results we do several measurements on (our) optical PUFs. These measurements allow us to conclude that a secrecy-rate of 0.3 bit/location is possible.

Unique Biometric Identifiers offer a very convenient way for human identification and authentication. In contrast to passwords they have hence the advantage that they can not be forgotten or lost. In order to set-up a biometric identification/authentication system, reference data have to be stored in a central database. As biometric identifiers are unique for a human being, the derived templates comprise unique, sensitive and therefore private information about a person. This is why many people are reluctant to accept a system based on biometric identification. Consequently, the stored templates have to be handled with care and protected against misuse [1, 2, 3, 4, 5, 6]. It is clear that techniques from cryptography can be used to achieve privacy. However, as biometric data are noisy, and cryptographic functions are by construction very sensitive to small changes in their input, and hence one can not apply those crypto techniques straightforwardly. In this paper we show the feasibility of the techniques developed in [5], [6] by applying them to experimental biometric data. As biometric identifier we have choosen the shape of the inner ear-canal, which is obtained by measuring the headphone-to-ear-canal Transfer Functions (HpTFs) which are known to be person dependent [7].