I was looking through my referrals this morning and came across a link from Podscope. The person had done a search for "technology and libraries" and Open Stacks #18 produced the only result. The apparent problem with this search is the presumption that this search tool operates in a Boolean manner. It does not.

Podscope is, at present, only capable of searching for words/phrases. It just so happens that I actually say "technology and libraries" in Open Stacks #18, so this searcher got a hit on their search. But I can't imagine that they were actually looking for someone to explicitly say those words in that order. Maybe I'm wrong.

A quick look at the FAQ reveals the following:

My complex search isn't returning any results. What's going on?

Supporting complex boolean searches is tricky when dealing with audio files. We do plan on providing boolean search capabilities, but we want some community feedback on how it should behave.

I don't see why Podscope would think Boolean is tricky. Once you've transcribed the audio, isn't it like any other text? Podcast search cousin Podzinger seems to treat all multi-term searches with a Boolean AND, unless quotes are used, which is more in-line with searcher expectations.

But here's something that's genuinely tricky: getting accurate transcription results. Automated speech-to-text transcription technology is shockingly mediocre.

During my presentation at CiL, I played a sample Podscope result on the search term "libraries" and, wouldn't you know it, no mention of "libraries" in the clip. Most folks in the audience were naturally expecting the word libraries and therefore totally missed the word "Blackberries" that was erroneously transcribed.

Podzinger shows keyword-in-context in their results, so you can get a feel for just how poor the technology is. A sample result for "libraries" reads as follows: "expensive business and that Contribution double life will be economy by libraries like good bet that there's certainly"

All this to say that, while these automated speech-to-text technologies are interesting and fun, you have to understand their limitations before relying on them as a search tool. Added bonus tip: Until the technology improves, use fewer and shorter search terms/phrases to minimize the likelihood of transcription error masking the results you're looking for.