Dhany Arifianto

Dual Parameters for Voiced-Unvoiced Speech Signal Determination

This paper describes the application of the notion of instantaneous frequency amplitude spectrum (IFAS) to discriminate voiced and unvoiced segment of speech signal. The classification procedures of speech signal into voiced and unvoiced is determined by using harmonicity measure acquired after evaluating instantaneous frequency amplitude spectrum. For accuracy improvement, we use secondary parameter during transition from voiced-to-unvoiced and unvoiced-to-unvoiced to confirm the voiced area estimated by IFAS. Entropy of magnitude spectrum and instantaneous power are considered in this investigation. The performance of the dual method is compared to single thresholding using IFAS and also ESPS, AMDF and TEMPO to demonstrate its effectiveness.

Source separation using independent component analysis techniques for machine fault detection in the presence of background noise

Early and accurate fault detection on a machine is crucial in preventive maintenance in order to prevent accidents that may be catastrophe to the user. However, direct measurement using vibrometer is common usage which may impractical in an industry that uses many rotating machines. In this paper, we evaluate the independent component analysis techniques which is time-, frequency-domain and multistage ICA for remote condition monitoring by analyzing sound emitted from the machines. We used electrical water pumps with normal, and intentionally introduced faults to the pumps with unbalanced, misaligned and bearing defect. These machines worked simultaneously and then recorded in an anechoic chamber to obtain the baseline data and then at an open area to simulate the real plant situation. We assumed that the sounds mixed convolutively at which required microphone array as sensor as an interface before separation. The results suggest that the proposed technique performed accurately in mean-square-error (MSE) sense. This implies that the proposed technique may be suitable for further implementation in real plant setting with adverse environment to replace current technique using direct measurement.

Underwater sound propagation characteristics at mini underwater test tank with varied salinity and temperature

In this paper, we report the development of water tank for underwater acoustic communication. Compared to most of underwater test tank, the tank may be considered as mini in term of its size. However, the test tank is intended to serve as controlled environment to develop a theoretical model and its validation experimentally before full-scale test at sea. We present on how to analyze the physical characteristics of the test tank with respect to salinity and temperature affected to speed and pattern of propagation waves. We designed the test tank made of 12mm thick of glass with dimension 2 × 1 × 1 in meter, then we placed corrugated sponges inside the walls to reduce the echo from signal that we generated. The measurements were conducted by placing array of hydrophones and analyze the propagating waves in fresh water with variation temperature 15°C, 20°C, 25°C and 30°C to simulate depth. In term of salinity we fixed at room temperature 30°C and varied the salinity of 3.1%, 3.2% 3.3%, 3.4% and 3.5%, respectively. The results showed that the sound propagation characteristics are in agreement with previous studies. This may suggest that the test tank may serve as small scale underwater acoustic experiment with controlled environment.

HMM-based Indonesian speech synthesis system with declarative and question sentences intonation

In this paper, we present a result of HMM-based speech synthesis system applied to Indonesian with prosody information in declarative and question sentences. The purpose is to observe the speech quality of synthesized speech from declarative to question sentences, conversely. Variation is given in kind of sentences and training data amount. We use 44, 72, 116, 450, 929 and 1379 training data. The result were evaluated by objective and subjective test. In objective test using MCD method earn the best score for question sentences with score 4.32 in 450 training data. Then for declarative sentences with score 5.13 in 929 training data. Subjective test with DMOS method obtain naturalness for declarative and question sentences with score 3.53 and 3.36 in 1379 training sentences respectively. The result shows that “degradation speech is slightly annoying”, it is possibly caused by poor F0 estimates.

Speech disorder analysis using time-varying autoregressive

In this paper, perceivable hoarseness classification of uttered speech is presented to determine degree of severity. Fundamental frequency of speech signal is estimated by using instantaneous frequency based and autocorrelation method. Degree of severity of the patients, in particular who are suffered from vocal nodule, are classified by observing the fundamental frequency distortion and aperiodicity of sustain vowel utterance. To add more information, we also observe behavior of the coefficient of autoregressive model with respect to time and frequency.

Just noticeable difference of masker to enhance privacy in an open-plan office

If a background conversation (masker) is easy to follow, then a worker at a workstation in an open-plan office will be distracted and annoyed. The aim of this study was if the mixture of the background conversations statistical distribution close to Gaussianity, then regardless the loudness level, the intelligibility will be decreased. On the other hand, the privacy at the workstation will be increased due to the loudness level of the background conversation. To assess the privacy level in a simulated open-plan office, we propose Just Noticeable Difference (JND) of the masker. The measurement was conducted in two workstations laboratory with 64 square meters for each workstation in which three conditions (male-female, all male, and all female speakers) of babble noise was built in one of them. The other workstation was simulated with single speech sound to observe speech privacy in the presence of the masker. We used objective measures to assess the intelligibility and the privacy of the workstation. The results suggest that the saliency of the masker depends on the fundamental frequency difference (dominant speaker). The higher saliency of the masker will cause the lower of the privacy of the other workstation.If a background conversation (masker) is easy to follow, then a worker at a workstation in an open-plan office will be distracted and annoyed. The aim of this study was if the mixture of the background conversations statistical distribution close to Gaussianity, then regardless the loudness level, the intelligibility will be decreased. On the other hand, the privacy at the workstation will be increased due to the loudness level of the background conversation. To assess the privacy level in a simulated open-plan office, we propose Just Noticeable Difference (JND) of the masker. The measurement was conducted in two workstations laboratory with 64 square meters for each workstation in which three conditions (male-female, all male, and all female speakers) of babble noise was built in one of them. The other workstation was simulated with single speech sound to observe speech privacy in the presence of the masker. We used objective measures to assess the intelligibility and the privacy of the workstation. The ...

Fundamental frequency evaluation of infant crying

Infant crying provides importance information about the babys physical and physiological conditions, such as health, gender and emotions. The anatomy of infants glottal causes different fundamental frequency (F0) production in its cry sound. It is generally accepted that the assumption of an adult voice has quasi-stationary in short period (about 100 ms). However, infant cry has much shorter stationary period (about 5 ms), therefore it is necessary to evaluate several F0 extraction techniques, namely MB-formant tracking, STRAIGHT, YAAPT, and YIN. The results showed that STRAIGHT and YAAPT were successfully extracted F0 of infant crying with accurate voiced and unvoiced regions. The results also showed that span of the F0 was within the range from 190Hz to 600Hz. This suggests that the infant F0 is higher than those of adult. We also evaluated formant of infant cry and the Wavsurfer was more accurate than that of Mustafa-Bruce Formant Tracker.

Enhanced harmonics for music appreciation on cochlear implant

The temporal fine structure in music consists of harmonics which is important for music appreciation. The lack of this temporal fine structure in the most used cochlear implant encoding strategy, contributes their lack of the music appreciation. The purpose of this paper is to enhance the harmonics for music appreciation on cochlear implant user. In the first experiment we processed the prior music signal with voice coder (vocoder). During prior processing, the harmonics of original music was degraded. In the second experiment, we used well known signal enhancement, Frequency-Amplitude-Modulation-Encoding (FAME), to enhance the harmonics of prior music. Objective test was conducted to measure the quality of music relative to original music using log spectral distance (LSD). The music quality between the prior and after enhancement are 2,3 and 2,1, respectively. The higher value of LSD means that the enhanced music got worsens. We also conducted subjective listening test from two subjects, 10 musicians and 10 non-musicians, whose their music appreciation is decreasing after the music was enhanced by FAME. It reveals that FAME strategy is not useful to enhance harmonics in signal processing of cochlear implant.

Data protection using interaural quantified-phase steganography on stereo audio signals

Internet is rapidly growing because it offers the easier and interconnected digital lifestyle. The growth also brings data security threats. Data are used to be protected by encryption. Encryption morphs the data into unreadable series of code which obey a certain rule. But the way encryption did to the data make the existence of precious data is exposed to danger. Steganography which differ from cryptography offers a stealth data protection without make any suspicion to the existence of data. Steganography hides the data through images, audio files, TCP/IP, etc. Phase encoded steganography is a robust information hiding method on audio file. It is because the ability to deceive human auditory system perception. Exploiting the insensitivity of human auditory system to distinguish a slight phase difference, interaural phase encoding sound modification turn out to be undetectable. Utilizing the base-4 and hexadecimal code number basis, instead of binary, embedded data capacity can grow significantly. At 33 dB SNR, it is possible to embed 1500 bit data per second. The higher code number base and the higher number of segmentation, the more minimum modification would happen on time domain. These base-4 and hexadecimal encoding are quite robust against attack based on amplification and filtering.

Localization of underwater moving sound source based on time delay estimation using hydrophone array

Signal and noise of an underwater moving sound source is used to track the azimuth of a target. Uniform linear array with four hydrophones is used to detect azimuth of target by obtain the time delay information to get azimuth information. Success rate of time delay estimation influenced by characteristics of sound propagation like reflection, reverberation, etc. Experiment in real environment was done to analyze performance of the cross correlation (CC) and generalized cross correlation with the phase transform (PHAT) weighting to estimate time delay between two signal. The simulation done by convolute two signal that has been given time delay and impulse response of the medium test. Then the time delay of two signal estimated by CC and PHAT algorithm in Matlab in the various SNR. Then the algorithm tested in a pool to detect stationary and moving position of sound source. Result of the simulation and experiment in real environment shown that PHAT better than CC. The best azimuth tracking achieved by using PHAT algorithm with error of 0 - 9.48 degree in stationary position. In moving sound experiments, tracking the bearing and azimuth of the mini vessel (sound source) can be done by time delay estimation using PHAT.