AS THE use of audio and video become increasingly common on investor relations websites and archives become deeper, IR website managers must consider ways to ensure that investors can easily find information in these formats.
The problem until now has been that information in audio and video is invisible to search engines, which rely on text to find matches to users’ search queries. Basic meta content — titles, descriptions and tags — and transcripts have traditionally provided the only way around this problem.
Two years ago, Danske Bank introduced a searchable webcast portal that relies of the company associating keywords with each webcast, but this is not always a good experience for users. Meanwhile, some high-end solutions — such as that used by Thomson Reuters’ StreetEvents service — have provided links between transcripts and time stamps in audio and video recordings, but this has been a slow and expensive method.
|Danske Bank uses meta tags to make it’s webcasts searchable. However, this is a less precise method of indexing multimedia.|
Now several companies are using advanced technologies to make video easier to search. In September 2008, a company called VideoSurf unveiled a new video search technology that analyzes the images in videos to identify people and other relevant information. The company is still in Beta, but from what we’ve seen of it so far the new technology leapfrogs any existing products on the market or under development.
Recently, Google began a limited trial of a speech recognition engine that “listens” to and transcribes the soundtrack of political videos on its YouTube service. The transcript is then synchronized with the video timeline so that users can jump to the exact place in the video where their search terms are mentioned. For example, search the political videos for the phrase “executive compensation” and the places in the video where the phrase is used is shown in the video timeline.
Google reports that it usually takes less than a few hours for a video to be transcribed and made searchable after it has been published on YouTube. The company admits that the technology is not 100% accurate but is useful for most purposes. It says “speech recognition is a difficult problem that hasn’t yet been completely solved, but we’re constantly working to refine our algorithms and improve the accuracy and relevance of these transcribed results.”
|Google is using speech recognition to make video searchable. In the above example, we’ve searched political speeches for mentions of the term “executive compensation.”|
Several other companies are using speech recognition engines to provide services and products to web publishers. They include Blinkx, which was spun off from Autonomy and developed its technology under government security contracts. The BBC uses the technology to search and organize its archives.
Similarly, EveryZing, formerly known as Podzinger, was spun off from national security contractor BBN. It provides two main services. ezSearch is a site search utility that indexes multimedia content as well as normal web pages. And ezSEO is a speech recognition engine that wraps audio and video in full-text meta data that can be indexed by search engines.
A newcomer to the field is Delve, which provides a technology similar to Google’s that lets users find keywords inside of video content.
Later this year, Adobe plans to include a speech-to-text engine in its Adobe Premier video editing software. This will enable easier editing of video for producers and the ability for web users to search inside video and audio that is published to the web.
This article originally appeared in the 2008, vol. 3 issue of Online IR Trends. Get your own subscription today.