Blog: Richard Hackathorn

« IBM IOD/Gold - Sensors Galore | Main | IBM IOD/Gold - Tom Davenport »

IBM IOD/Gold - Speech Analytics

The next Gold session is about Speech Analytics: Extracting Business Insight from the Voice of the Customer by David Nahamoo, IBM Fellow, Speech CTO and Business Strategist. He set the context for speech analytics with a figure of the data spectrum from structured to unstructured (which I thought that this was a nice concise illustration).

The market for speech analytics is currently driven by application for: government intelligence, media analytics, and contact (call) centers. David concentrated on the contact centers, where the goals were: quality assurance monitoring, customer satisfaction, increasing sales, marketing effectiveness.

We had an interesting discussion of the quality (error rates) of speech mining. David shared a variety of speech sample to motivate the difficulties with accent, noisy backgrounds and the like. The technology for speech to text is getting closer to the quality of human transcription, which is in the 90% range. However, many commercial applications need only 50% accuracy, which allows processing at 5x real-time speed. In other words, 5 hours of call center recording can be processed in one hour.

My take on this. . . It is amazing the progress that speech translation technology has had in the last three years. We are only beginning to imagine the business potential of this technology.

[Blog stream from IBM IOD/Gold October 2008 is here]

Share: