Learn More
In Japanese, every content word has its own H/L pitch pattern when it is uttered isolatedly, called accent type. In a TTS system , this lexical information is usually stored in a dictionary and it is referred to for prosody generation. When converting a written sentence to speech, however, this lexical H/L pattern is often changed according to the context,(More)
In this paper, we introduce a novel framework for lossless/near-lossless (LS/NLS) color image coding assisted by an inverse demosaicing. Conventional frameworks are typically based on prediction (and quantization for NLS coding) followed by entropy coding, such as the JPEG-LS for bit rate saving. The approach of this work is totally different from the(More)
For developing a Japanese TTS system, it is important to realize a module of word accent sandhi adequately. Our previous works developed the module by using rules but it is a fact that the rule-based module produced an unignorable number of errors. In this report, accent labeling of a large-scale corpus was done by a single labeler and a corpus-based module(More)
We principally used two strategies for gene prediction and combined the results (see Supplementary Information for details). (1) Each read-pair of cDNA clones was mapped on the contigs using BLAST and putatively transcribed regions were determined by clustering the mapped pairs. (2) ORFs likely to encode a protein showing similarity to known proteins or(More)
  • 1