One of the more interesting uses of Intelligent Voice is in Case Management and eDiscovery.
Voice is one of the dark places that litigators fear to tread. It is expensive to transcribe, and expensive to listen to. Often the cost of this is disproportionate to the amount claimed and the other legal fees. And yet, the key piece of information may well have been let slip over the telephone. How best to balance cost, accuracy and speed?
Intelligent Voice is designed to work with larger volumes of voice data. Presented with, say, 100 hours of phone calls would be as quick and cheaper probably to get the calls transcribed, and we have a number of excellent providers who can do just that.
However, when you start to take large volumes of calls across multiple parties, we begin to make a lot more sense.
Look at the process flow below. This gives an example of how we work with an external case management and eDiscovery system such as Clearwell. We take the voice calls at one end, and produce something that can be ingested into your chosen platform at the other
Taking the various pieces in order:
- Speaker Separation: One of the more irritating parts of dealing with phone calls is how voice recording systems store and present their recordings. Generally, they are in mono, which is hard for a person to understand, let alone a computer. We apply biometric algorithms to the recordings to produce a “quasi-stereo” file that puts one speaker on one channel and one on the other. We can’t do this when people speak over each other (nor can anyone), which is why recording in stereo is always preferable.
- Speaker Identification: Often caller have no caller ID or have incorrect metadata (someone else picked up the phone), so we try to a) ensure that all known callers are correctly identified, and b) to group unknown callers, so that the 20%-40% of inbound calls that have no caller ID can be much better grouped and identified.
- Speech Recognition: Here we extract as much useful text as we can from the people speaking on the phone. Speech recognition is not perfect, and telephone speech recognition is even harder. However, we apply state of the art techniques to getting the best possible text we can in the shortest possible time.
- JumpTo(tm) Analysis: Unique to Intelligent Voice, we scan the transcripts to find interesting words and phrases so that you can quickly “Jump To” the relevant point in the conversation.
- Email Encapsulation: We then take the voice file, its bookmarks and text, as well as all of the metadata we have gathered, and encapsulate it into a format that can easily be digested into any eDiscovery or other text retrieval engine.
- Playback: The email is completely self-contained, and as the original file is contained within it, it can be played directly from any player (eg Windows Media Player). However, there is also a link to the “JumpTo” player, which gives enhanced bookmark and playback facilities