Podcastplayer.org news

Another intriguing idea. A speech-recognizing search engine that “listens” to podcasts and indexes the words. The site is full of bullish claims, but I think I’ll wait until I see it in action before jumping on the bandwagon.

Podscope is the Internet’s first spoken-word search engine for audio and video podcasts.

Theoretically, parsing words from something like a podcast should be a better deal than real-time speech-recognition. The software can take as long as it likes (within reason) trying different approaches to get a reasonable result. What worries me, though is the diverse nature of podcasters and podcast content.

All real-time speech-recognition systems that I’m aware of require some sort of “training", to get a grip on how the speaker uses even well-known words. Attempting to process an unknown podcast which may be in any language, in any accent, may be a mixture of voices, may have background music or chunks of non-spoken content seems a tall order.

My guess is that they will initially just “cherry pick” words that they are pretty sure about, and simply not index the rest. The trouble is that this is often the opposite of what’s needed when providing a searchable index. When searching you quickly learn that searching for rarer, more-specific words provides a better result; but these are just the kind of words that an automatic parser will lack the context to recognize.

Maybe they’ll get smart and support a wiki-style mass-participation system to allow anyone to correct words and feed back into teaching the system about hot ideas and specific podcasting styles.

Read more at: podscope - We’re listening. You’re searching