Using this speech based W matrix we develop the speech enhanced hybrid n- gram + LSA model. 3. Speech enhanced multi-span language model The block diagram of the proposed system for construction of matrix W is shown in Figure 1. ... text corpus of news bulletins or news wire corpora in Tamil language. ... to the domain of use, we used a bigram Compressed pattern vector w Closeness Template generation Mean patteml Normalised -2250- 1NTERSPEECH 2004 - 1CSLP.
|Title||:||Eigth International Conference on Spoken Language Processing|
|Author||:||International Speech Communication Association|