Browse Prior Art Database

Compression Method for Voice Preprocessing and Postprocessing

IP.com Disclosure Number: IPCOM000061796D
Original Publication Date: 1986-Sep-01
Included in the Prior Art Database: 2005-Mar-09
Document File: 2 page(s) / 14K

Publishing Venue

IBM

Related People

McDonald, BS: AUTHOR

Abstract

A method is described in which voice algorithm development is reduced by providing a common software interface with the analog hardware. This interface minimizes tuning time by providing a normalized input to the compression routine from all the different analog input devices. In addition, this 'preprocessing' block provides a considerable amount of common function that is required by the voice partition architecture. With appropriate definition the preprocessing block can help reduce the number of instructions that must be processed. In the record function, the preprocessing block provides a common interface to any compression algorithm.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 51% of the total text.

Page 1 of 2

Compression Method for Voice Preprocessing and Postprocessing

A method is described in which voice algorithm development is reduced by providing a common software interface with the analog hardware. This interface minimizes tuning time by providing a normalized input to the compression routine from all the different analog input devices. In addition, this 'preprocessing' block provides a considerable amount of common function that is required by the voice partition architecture. With appropriate definition the preprocessing block can help reduce the number of instructions that must be processed. In the record function, the preprocessing block provides a common interface to any compression algorithm. It will normalize the input samples, identify silence and super-compress it, post interrupts during silence periods at a rate determined by the user and provide a volume level feedback to the user, (similar to VU meter on a tape recorder). Any compression routine can then be used and meet most of the requirements of the architecture. Normalization of the input signal is used to minimize the calculation error of the compression routines. To minimize the round-off error of calculations, as many bits as possible should be utilized. The input range of the voice from different devices can vary by 30 dB. Therefore, the input signal is normalized to a common reference point to minimize the calculation error due to the input variation. The AGC is the primary element in the implementation of this design. The AGC is an automatic gain control module whose primary function is to normalize samples to a reference level. The AGC is designed and tuned to optimize its performance for a voice signal. One of the natural fallouts of the design is a parameter that represents the relative energy of the incoming signal. This parameter allows the determination of silence and volume level of the input signal before the samples are normalized. Once the silence is identified, it is super-compressed from the input samples. This is performed by placing a unique silence header and count in the data stream where the count is proportional to the length of the silence period. The term "super-compression" is used to differentiate the silence compression from the normal algorithm compression. The silence compression stores minutes of silence in just a few words of RAM since there is no useful information in a silent signal. Silence can then be reproduced from the count when the message is replayed. Identifying and super-compressing silence in the preprocessing block results in a substantial performance improvement. This performance improvement is obtained whe...