Broadcast news transcription is one of the hardest tasks in speech recognition because broadcast speech signals have much variability in speech quality, channel and background conditions. We developed a Korean broadcast news speech recognizer. We used a morpheme-based dictionary and a language model to reduce the out-of·vocabulary (OOV) rate. We concatenated the original morpheme pairs of short length or high frequency in order to reduce insertion and deletion errors due to short morphemes. We used a lexicon with multiple pronunciations to reflect inter-morpheme pronunciation variations without severe modification of the search tree. By using the merged morpheme as recognition units, we achieved the OOV rate of 1.7% comparable to European languages with 64k vocabulary. We implemented a hidden Markov model-based recognizer with vocal tract length normalization and online speaker adaptation by maximum likelihood linear regression. Experimental results showed that the recognizer yielded 21.8% morpheme error rate for anchor speech and 31.6% for mostly noisy reporter speech.

Two improvements are discussed in measurement of the complex Young's modulus of the acoustic materials by using the transfer function method. It is found that the accelerometer misalignment might result in the severe measurement error, particularly in high frequency range. The supporting structure is modified to attach the upper and lower accelerometers along the vertical axis. Secondly, the method fur solving the equation associated with wave model is described. The solution of the lumped mass-spring model is chosen as the starting value for low frequency range, while in the mid and high frequency, the solution to the previous frequency step is used as the initial values. Measurements are done for hard and soft rubber specimens. It is shown that the erroneous peaks in the transfer function, due to the measurement error, cause highly incorrect Young's modulus and loss factors.

This paper discusses the development of an acoustic vector channel simulator for the performance analysis of an acoustic digital communication system. The channel simulator consists of transmission module, acoustic channel model, receiver, beamformer, and adaptive equalizer. The source signal (QPSK) is generated by the specified parameters. The transmitted signal generates multipath signals which have a different delay, amplitude and doppler frequency. The paper presents in details the approach to the performance analysis of an acoustic digital communication system according to the antenna structure and the various baseband signal processing techniques.

We report on the development of a fiber optic hydrophone consists of a sensing Michelson interferometer and a compensating Mach-Zehnder interferometer for optical path length compensation. The double interferometer configuration has the following advantages: the hydrophone can be made more small; a laser source with a relatively short coherence length can be used; and the compensating interferometer can be located near the signal processing electronics, far away from the sensing interferometer and noise introduced by reference arm can be greatly reduced. The performance of the hydrophone is evaluated experimentally by immersing the sensing interferometer in a water tank to detect underwater acoustic signals generated by an acoustic wave projector. Experimental results show that over the frequency range of 1 to 4 kHz, the hydrophone has an almost flat response with an average normalized sensitivity of -302 dB re 1/ μ Pa.

This study aims at investigating the relationship between voice and the image information carried within the voice. Whenever we hear somebody talking, we form a mental image of the speaker. Is it accurate? Is there a relationship between the voice and the image triggered by the voice? To answer these questions, speech samples form 8 males and 8 females were recorded. Two photos were taken for each speaker: the whole body photo (W) with physical characteristics present, and the face close-ups (F) without much physical details revealed. 361 subjects were asked to match the voices with the corresponding photos. The results showed that 5 males and 5 f3males (with W) and 2 males and 4 females (with F) were correctly identified. More interestingly, however, even in the mismatches, there was a strong tendency for participants to agree on which voice should correspond to which photo. The participants also agreed much more readily on their favorite voice than on their favorite photo. It seems voice does carry certain information about the physical characteristics of the speaker in a consistent manner. These findings have some bearings on understanding the mechanism of speech production and perception as well as on improving speech technology.

In this paper, a computer simulation method is presented far the dynamic analysis of a hydraulic engine mount system. The hydraulic engine mount system controls the damping characteristics using the viscosity of fluid flow. The complex stiffnesses of the main rubber for the hydraulic engine mount system are computed using a finite element analysis. The equations of motion considering the parameters of the hydraulic engine mount system are derived. To investigate the effects of the hydraulic engine mount system, the computer simulation running over a typical rough road is carried out using a vehicle dynamic model. These results are compared with those of the conventional rubber mount system.