Manya's thesis: "Combining voice capture with business intelligence, analytics, and text mining provides valuable customer intelligence for marketing and competitive intelligence business functions." Hers is a marketing paper, but it's motivated by significant technical and business challenges. I've looked at the issues myself and have planned to research them and couldn't agree with her more. (That's why we plan significant coverage of speech mining and audio analytics at the Text Analytics Summit, which I chair, in 2009.)As in dealing with text, businesses typically try search, followed by manual review of recordings, as a first shot at taming massive volumes of captured audio. Manya characterizes this approach as "backward-looking." More-comprehensive, automated treatment, starting with transcription to text and application of data mining techniques, facilitates the creation of forward-looking predictive models. Further, non-transcribed information captured in the audio signal such as
- call length,
- emotion and stress indicators,
- and silences, holds, and transfers,
Manya advocates use of phonetic indexing that "transforms the captured audio signal into a sequence of phonemes or sounds" to support audio search. She goes on to describe the use of SAS Text Miner, in conjunction with audio output captured by systems from vendors including CallMiner, NICE Systems, and Witness Systems (Verint) — Autonomy and Nexidia are two others in this domain — and how the software and SAS Enterprise Miner are used to support call-center analytics at SAS customer MSNTV. Do note that roughly equivalent capabilities and services are offered by SAS analytics rivals including Clarabridge, IBM, SPSS, and Expert System, which partners with speech-technology provider Loquendo. (I write an editorially independent column for Clarabridge's quarterly newsletter, and IBM and SPSS sponsored my recent report on Voice of the Customer Text Analytics.) Other text-analytics vendors in the customer experience management/CRM space report exploring audio-mining possibilities.
Hardware that records and stores massive volumes of audio data is cheap, and it's being used by consumers for personal purposes, by corporations for customer support, and by law-enforcement and intelligence agencies. Customer service isn't the only corporate imperative; e-discovery rules require retention and the ability to classify and produce electronic records including audio. The mandates are out there, and technology, as surveyed in Manya Mayes's voice-mining paper, is filling an essential role in responding to them.Manya Mayes of SAS has written a helpful introductory paper on audio analytics, "Tune into the Voice of Your Customer with Voice Mining." The technology and techniques described have applications for e-discovery, intelligence, and rich-media search. Given coverage of distinctive characteristics of speech and of analytical concerns that include BI integration, Manya's paper merits a look for anyone who works with audio data.