This paper proposes a method to develop a complete speech compression system using Devanagari script. Any speech can be considered as a concatenation of basic sounds, called 'phonemes'. Most of this basic set is the same for all... more
This paper proposes a method to develop a complete speech compression system using Devanagari script. Any speech can be considered as a concatenation of basic sounds, called 'phonemes'. Most of this basic set is the same for all languages, with a few language-specific additions. If this set can be identified, then we can use these phonemes as the fundamental building block for all speech. Words in any language can then be represented using a Devanagari Transform, which expresses them as a combination of Devanagari phonemes. This lends the process its language-independence. This approach involves developing a phonetic recognizer which segments the words into a set of phoneme units, using Hidden Markov Models. These recognized units are mapped onto a labeled phonetic database containing about 4000 phoneme units. The uttered speech can then be expressed using only the pointers to these entries. This results in compression by a factor of up to 1000. The speech is reconstructed b...