Efficient compression of vrom messages for telephone answering devices

Abstract

A telephone answering device including two separate coders, a first coder for encoding/decoding fixed voice prompts spoken by a single speaker, and a second coder for encoding/decoding incoming and outgoing voice messages spoken by multiple speakers. The first coder uses a first set of codebooks trained based on a first set of utterances spoken by a single speaker, while the second coder uses a second set of codebooks trained based on a second set of utterances spoken by multiple speakers. Because the first set of utterances is significantly smaller in size than the second set of utterances, and the range of pitch period is significantly smaller in size for the first set of utterances spoken by a single speaker in comparison to that of the second set of utterances spoken by multiple speakers, the size of the first set of codebooks is significantly reduced relative to the size of the second set of codebooks. As a result, the fixed voice prompt messages may be compressed at a lower bit rate with a relatively high quality of encoding, thereby optimizing the codebook and reducing the amount of necessary memory capacity for storing the encoded fixed voice prompts. The memory required for the encoded first voice prompts is so small that they can be stored in a low cost DSP ROM.