In reinforcement learning, large state and action spaces make the estimation of value functions impractical, so a value function is often represented as a linear combination of basis functions whose linear coefficients constitute parameters to be estimated. However, preparing basis functions requires a certain amount of prior knowledge and is, in general, a difficult task. To overcome this difficulty, an adaptive basis function construction technique has been proposed by Keller recently, but it requires excessive computational cost. We propose an efficient approach to this difficulty, in which the problem of approximating the value function is decomposed into a number of subproblems, each of which can be solved with small computational cost. Computer experiments show that the CPU time needed by our method is much smaller than that by the existing method.

This paper describes an ITU-T G.711 embedded wideband speech coder, submitted as a candidate to the ITU-T G.711 wideband extension standardization qualification phase. The codec generates a bitstream comprised of three layers: a G.711 core layer with noise shaping, a time-domain weighted vector quantized narrowband enhancement layer, and an MDCT-based higher band enhancement layer. Through subjective evaluations, the coder was found to meet all tested requirements and objectives set in terms of reference, with a low computational complexity at 10 WMOPS.

Reinforcement learning (RL) methods based on least-squares temporal difference (LSTD) have been developed recently and have shown good practical performance. However, the quality of their estimation has not been well elucidated. In this article, we discuss LSTD-based policy evaluation from the new view-point of semiparametric statistical inference. In fact, the estimator can be obtained from a particular estimating function which guarantees its convergence to the true value asymptotically, without specifying a model of the environment. Based on these observations, we 1) analyze the asymptotic variance of an LSTD-based estimator, 2) derive the optimal estimating function with the minimum asymptotic estimation variance, and 3) derive a suboptimal estimator to reduce the computational burden in obtaining the optimal estimating function.

This paper proposes a lossless audio coding scheme with hierarchical scalability in terms of sampling rate and amplitude resolution. A single bit stream contains hierarchical information that can generate waveforms ranging from 96 kHz with 24-bit amplitude resolution through lower sampling/resolution lossless waveforms to a highly compressed lossy one created using an MPEG-4 audio coder. This bit stream structure enables dynamic rate control and hierarchical multicasting based on a simple priority control of the IP packets. These functions will be useful for high-quality archiving and broadband streaming for various types of networks and terminal equipment.

This paper proposes a lossless scalable audio coding scheme and quality enhancement processing at the decoder to compensate for some missing scalable units of information. The bit rate scalability is achieved by combining high-compression coding, such as MPEG-4, and horizontal bit slicing of the PCM-coded error signal between the original waveform and the locally reconstructed MPEG-4 signal. The horizontally sliced stream may be transported through an IP network with priority. Even if some units are missing at the decoder, reasonable quality waveform can be reconstructed by means of post-processing using some side information. This scheme enables graceful degradation by supporting lossless, near lossless, and high-compression coding within a single scalable framework, and is useful for narrowband to broadband audio streaming.

The ISO/IEC MPEG-4 audio standard includes the TwinVQ encoding tool. This tool is suitable for low-bit-rate general audio coding, but a drawback is the computational complexity of the encoder. To develop a faster TwinVQ encoder, new fast vector quantization algorithms - area localized pre-selection and hit zone masking - are introduced. These algorithms exploit the pre- and main-selection procedure scheme of the conjugate structure vector quantization which is used in the TwinVQ. The improvement is evaluated by measuring the encoding speed and the sound quality of reproduction.

This paper proposes a lossless audio coding scheme making use of high-compression lossy coding such as the ISO/IEC MPEG-4 audio standard. The encoding process consists of lossy compression, bit slice lossless data conversion, and lossless coding. The lossy compression significantly reduces the amplitude of the error signal between the input signal and the reconstructed signal. In addition, bit slice data conversion makes the lossless coding more efficient. As a result of preliminary experiment, a total file size can be reduced to 70-50% of the original size, which is significantly more efficient than conventional universal lossless coding schemes such as "gzip". The resultant compressed data is useful for various applications, since it has bit rate scalability and can be used as a highly compressed signal as well as a lossless signal.

This paper proposes a combined design for an error robust and scalable audio coder based on TwinVQ (transform domain weighted interleave vector quantization) object type and EP (error protection) tools defined in MPEG-4/Audio. By combining the inherent error-robustness of TwinVQ, the un-equal error-protection scheme, and the hierarchical scalable coding structure, the distortion is confirmed to be very minor, even in high-error channel conditions at bit rates of 16-32 kbit/s. These structures offer a flexible coding framework which can be used for various types of network/storage environment and applications.

A large-vocabulary continuous-speech recognition (LVCSR) system was developed and evaluated. To evaluate the system, a Japanese business-newspaper speech corpus was designed and recorded. The corpus was designed so that is can be used for Japanese LVCSR research in the same way that the Wall Street Journal (WSJ) corpus, for example, is used for English LVCSR research. Since Japanese sentences are written without spaces between words, a morphological analysis was introduced to segment sentences into words so that word n-gram language models could be used. To enable the use of detailed word n-gram (n/spl ges/3) language models, a two-pass decoding strategy was applied. Context-dependent (CD) phone models and word trigram language models reduced the word error rate from 80.2% to 10.1% (an error reduction of about 88%). This result shows that CD phoneme modeling and word trigram language models can be used effectively in Japanese LVCSR.