Infant cry signal is a biomedical acoustic signal that is usually high-pitched. Infant cry is his only means of communication, production of which involves significant changes in its characteristics. Possibly rapid changes in its short-time segments carry information about the cause of cry, that a mother can perceive. The instantaneous fundamental frequency (Fo) of infant cries is much higher than for adults, necessitating different signal processing methods for infant cry signals' analyses. In this paper, the production characteristics of infant cry signals due to pain vs. discomfort are discriminated, exploring features that may help towards developing automated diagnostic systems. Production features the F 0 , strength of excitation (SoE) and signal energy (E) are used, for analysing ‘Infant Cry Signals Database’ collected for the study. The excitation source characteristics are derived using a modified zero-frequency filtering method. Spectrograms of the acoustic signal and the excitation source characteristics are compared to validate the changes in the features. Larger fluctuations in the features F 0 , SoE and E are observed for infant cry due to pain, than due to discomfort.

Discriminating features of infant cry acoustic signal for automated detection of cause of crying

Asphyxia is a respiratory injury that leads to a serious damage for infants. Early detection of asphyxia using Artificially Intelligent technology helps in reducing infant mortality rate when compared to traditional medical diagnosis, which is time consuming. In this paper, we propose a novel method through generating weighted prosodic features combined with acoustic features to form a merged feature matrix to classify asphyxiated baby crying effectively. The weights of the prosodic features are trained at the frame level with labeled data and can be optimized using deep learning approach with neural networks. The novel merged feature matrix is established with both acoustic and weighted prosodic features. The matrix has good ability to capture the diversity of variations within infant cries, especially for asphyxiated samples. Our method has the benefits of keeping the robustness and resolution of the classification model simultaneously. The effectiveness of this approach is evaluated on Baby Chillanto Database. Our method yields a significant reduction of 3.11%, 3.23%, and 1.43% absolute classification error rate compared with the results using single acoustic features, single prosodic features, and both acoustic and prosodic features, respectively. The testing accuracy in our method reaches 96.74%, which outperforms all other related studies on asphyxiated baby crying classification.

Deep Learning for Asphyxiated Infant Cry Classification Based on Acoustic Features and Weighted Prosodic Features

A maximum likelihood (ML) estimator for digital sequences disturbed by Gaussian noise, intersymbol interference (ISI) and interchannel interference (ICI) is derived. It is shown that the sampled outputs of the multiple matched filter (MMF) form a set of sufficient statistics for estimating the input vector sequence. Two ML vector sequence estimation algorithms are presented. One makes use of the sampled output data of the multiple whitened matched filter and is called the vector Viterbi algorithm. The other one is modification of the vector Viterbi algorithm and uses directly the sampled output of the MMF. It appears that, under a certain condition, the error performance is asymptotically as good as if both ISI and ICI were absent.

Maximum likelihood receiver for multiple channel transmission systems

Cry is a means of communication for an infant. Infant cry signal is usually perceived as a high-pitched sound. Intuitively, significant changes seem to occur in the production source characteristics of cry sounds. Since the instantaneous fundamental frequency (F0) of infant cry is much higher than for adults and changes rapidly, the signal processing methods that work well for adults may fail in analyzing these signals. Hence, in this paper, we derive the excitation source features F0 and strength of excitation (SoE) using a recently proposed modified zerofrequency filtering method. Changes in the production characteristics of acoustic signals of infant cries due to pain and discomfort are examined using the features F0, SoE and signal energy. These changes are validated by visually comparing their spectrograms with the spectrograms of the acoustic signals. Effectiveness of these discriminating features is examined for different pain/discomfort cry sounds pairs in an ‘Infant Cry Signals Database (IIIT-S ICSD)’, especially collected for this study. Fluctuations in the features F0, SoE and energy are observed to be larger in the case of infant cry due to pain, than for discomfort. These features can help in developing further the clinical assistive technologies for discriminating different infant cry types and initiating the remedial measures automatically.

Discriminating the Infant Cry Sounds Due to Pain vs. Discomfort Towards Assisted Clinical Diagnosis

Due to recent crowdedness in unlicensed spectrum, a new technology is introduced which allows unlicensed users, also known as Secondary Users (SUs), to dynamically access licensed spectrum whenever they are not used by their licensed users, also known as Primary Users (PUs). Routing in Cognitive Radio Networks (CRNs) is not as in traditional routing of wireless networks, because it is mandatory for SUs to periodically sense licensed spectrum, channels availability changes over time, knowledge of tolerated interference by PUs. In this paper, a novel routing discovery technique is proposed to find a primary path, if exist, which is immuned to one PU, such that at most one SU fails (must back off transmission) when a PU becomes active again. Also, a backup path can be discovered using the same technique with the condition that it is channel-link disjoint from the primary path. The problem is modeled using a multi-layer graph where each layer corresponds to a channel in a network. The proposed strategy reduces number of required channel-links maintenance to two, if failed due to one PU activity. As a result, SUs’ energy consumption is reduced when re-establishing route connectivity, and therefore, network life time is prolonged.

Finding an Immuned Path against Single Primary User Activity in Cognitive Radio Networks

Infant crying comprises a rhythmic pattern of cry sounds and inhalation. Unlike in adults, crying is the only means of communication for an infant. Most signal processing tools that work well for adults are not adequate in the case of infant cry sounds. Hence there is a need to develop methods for extracting features from these sounds, for better understanding. This paper describes a database collected for the analysis of infant cries vis-a-vis their causes, using spectrograms. The fundamental frequency, limited to adults, can go much higher in the case of infant cries, along with rapid changes in F0. Signal processing methods like autocorrelation and linear prediction analysis are used for analyzing the infant cry sounds and extract features like fundamental frequency, energy etc. Spectrograms providing the ground truth and information about the fundamental frequency with harmonics are examined in this preliminary analysis. An attempt is made to classify the infant cries into six categories such as pain, discomfort, ailments, emotional need for attention, hunger and cry due to manipulation.

Preliminary analysis of causes of infant cry

Infant cry is a biological signal through which an infant communicates with its care-giving environment. It also contains valuable information about the state of the infant. Infants produce this sound in response to a stimuli, which could be pain, discomfort, emotional need of attention, ailment, environmental factors or hunger/thirst. Signal processing methods that work well for adults are not adequate in the case of infants. In order to analyze the infant cry signals, these methods require some modifications. Signal processing methods such as short-time Fourier transform, auto-correlation and linear prediction analysis are modified and used. Features such as frame-energy and fundamental frequency are extracted from the cry signal. An Infant Cry and Causes (ICC) database especially collected for the study is used. Ground truth information about the fundamental frequency (F 0 ) is obtained using the spectrograms. The fluctuations in F 0 are examined using mean and standard deviation, for different causes of cry. The results about fluctuations in F 0 obtained from three different signal processing methods are compared. Objective is to classify the infant cry sounds into different categories of causes, based on the features derived from the infant cry signal. The results of the limited analysis are encouraging.

An investigation into classification of infant cries using modified signal processing methods

In this paper, we propose a Multi-user Multiple-Input Multiple-Output (MU-MIMO) based Cognitive Radio network where the unlicensed secondary uses (SU) can simultaneously use the spectrum which is used by a licensed primary user (PU). The interference caused by the secondary user to the primary user has been cancelled. This is done by using Block Diagonalization (BD) followed by QR-MRL in order to cancel the interference caused by PU antennas. Previous works, such as ZF (Zero Forcing)-BD and MMSE (Minimum Mean Squared Error)-BD are succeed to cancel the interference among multiple users. Here, the performance comparison has been made in terms of Bit Error Rate (BER) plots for ZF-BD, MMSE-BD and QR-MRL-BD. Also, these techniques have been compared with BER plot obtained when SUs interfere with the PUs. The results show that the BD method is able to cancel the interference caused by the SUs successfully and the QR-MRL method which is an ordered successive interference cancellation(OSIC) method effectively cancels the interference caused among user's antennas. A significant improvement in BER plot for QR-MRL-BD is observed. Also, a simulink model for ZF-BD has been discussed for future hardware implementation.

Naman Varma

Papers

Preliminary analysis of causes of infant cry

An investigation into classification of infant cries using modified signal processing methods

Interference cancellation in cognitive radio networks using block diagonalization