Voice Data Compression in Recording Solutions: A Blessing or a Curse?

Compression of voice data is widely used for digital recording solutions in analog and digital telephony environments to reduce the storage footprint for recorded audio. For VoIP, compression is required for low-bandwidth network links; however, recording solutions normally store the audio data in the same format as it is received over the network.

 

Compression reduces the bandwidth requirements for audio transmission and decreases the footprint for short-term storage and archiving. It may prove critical for call centers, but management must understand the technical process and limitations of compression prior to implementation.

 

TDM transmits audio PCM (pulse code modulation) encoded with 64 kbit/sec bit rate, which corresponds to the G.711 codec used in VoIP. One minute of audio of this quality consumes approximately 480 Kbyte of storage. The same audio will use only 60 Kbyte of storage when compressed with the common G.729 codec. The storage footprint for audio archiving directly impacts the cost for physical hardware (hard disks, NAS, SAN, DVD, etc.), and the processing time for transmission and storage/retrieval. It can also indirectly affect costs related to energy consumption, especially during peak hours.

 

Audio compression algorithms take advantage of limitations of the human sense of hearing and recurring sound patterns in human language (codebook technique). A limitation used since the introduction of telephony is the limited frequency band a human ear needs to “decode” language. With the exception of the G.722 high-quality audio codec, all current telephony systems utilize a frequency band between 300 Hz and 3,400 Hz. Frequencies above or below this range are cut off before transmission.

 

More sophisticated compression algorithms called CELP (code exited linear prediction) combine PCM and codebook compression techniques to achieve higher compression rates while maintaining acceptable speech quality. These methods include G.729 and iLBC, and are used for speech only, unlike the well known mp3 audio compression.

 

Audio compression is an irreversible process reducing the amount of data at the cost of audio quality. Though it is optimized for human hearing, compressed audio interferes with computer-based speech analytics and transcription, often to the point where they basically stop working when fed with compressed audio material.

 

While data compression can save a huge amount of storage space and speed up archiving and retrieval operations in call recording solutions, requirements for speech analytics or automated transcription must be carefully considered prior to its implementation.

 

Speech analytics offers special benefits in quality monitoring solutions and can improve agent performance through features such as emotion detection, keyword spotting and transcription. For organizations with a large volume of communications, it can automatically categorize recorded calls to identify the most interesting, critical and useful interactions among an otherwise unmanageable number of communications.

 

Audio compression should be carefully considered prior to implementation in light of its benefits and drawbacks. If conflicting applications are important, audio data compression should be avoided.