Categories
World Wide Web

Speech recognition -> Podcasts -> Podzinger

Podzinger is just what I was looking for! Podzinger uses speech recognition technologies to actually try and figure out the words in a podcast and then helps us to search within podcasts! Although not quite 100% accurate, it is quite impressive.

This can actually be used in a number of ways:

* Just search for keywords the way you do a normal search and get the podcasts of your choice. Podzinger actually provides RSS alerts for these keywords and so you get podcasts on the fly delivered to your favorite reader.

* I had recently written about the Problems with podcasts, where I had mentioned:

…there is an inherent problem with podcasts. They are not searchable. A typical podcast, for example, Slashdot Review contains many different news items. In this example, Slashdot review contains all the important stories published in Slashdot in that day.

In RSS, suppose I am not interested in reading a particular news item, I can just skip and read the next one.

Podzinger helps us with this.

Usually podcast publishers provide you with a description, which tells you what the podcast contains. Just use this to search in Podzinger and you can magically be transferred to the exact location where that particular item starts.

One problem however: Podzinger works only with IE 5.0+ with RealPlayer. (However for the sake of using this utility you can definitely go back and use that browser. 🙂 )

If you care about podcasts, you definitely should give it a try!

Categories
World Wide Web

Key-Value Tagging

The act of tagging consists of labelling objects with keywords [Wikipedia].
Tagging, the way it works now, is attaching separate keywords with
objects. Although we might attach multiple keywords with the same
object, the words are independent of each other (Don't argue that the
words are related in the sense of tag clusters. Let me get to the
point).

In its present form, tagging no wonder has created a revolution. But
would it not be more useful if tagging were in the form of key-value
pairs as well. I should have an option of either tagging objects with
single words (as it works now), or with key-value pairs.

How would this help? I had written about Problems
with Podcasts
sometime back. Now consider a model in which I
could not only have skip-points which mention where a particular topic
starts, but also what these topics are and my own comments on it.

If you compare a single podcast to a set of blog entries, 'key-value'
tagging could be compared to comments to a single blog entry. It would
look somewhat like this:

<skippoint>
 <time>0.24.29</time>
 <comment>This is where the speaker talks about Google's WebOS initiative.</comment>
</skippoint>

Although this can be done using XML so easily, an end user would not
like writing XML code. So a simple interface could be provided where
the user writes the time and the comment and this is clubbed with the
podcast and can be accessed anywhere on the web. Further, the user
could add any information, for example, the name of the speaker
(example, speaker=Gautham) or the location where the podcast was
created (example, location=Bangalore).

And just like tags, nothing is pre-defined. The user can add just about
any 'key-value' tags to any object. Again, as I keep mentioning, RDF
has solutions to these. But 'Keep It Stupidly Simple' is how the web
works. So be it. 🙂

I have been talking about Tag evolution here.

Categories
World Wide Web

Problems with Podcasts

Podcasts are the new buzz thing in the WWW. While RSS provides a mechanism to subscribe to textual feeds, Podcasts help in subscribing to audio/video content. So, instead of those small orange bars, you will now see colorful iTunes images or Odeo images.

However there is an inherent problem with podcasts. They are not searchable. A typical podcast, for example, Slashdot Review contains many different news items. In this example, Slashdot review contains all the important stories published in Slashdot in that day.

In RSS, suppose I am not interested in reading a particular news item, I can just skip and read the next one. But in Podcasts, since all the news items are aggregated together into a single audio feed, we are not able to skip certain items.

However considering the fact that Podcasts are still in their infancy, we can expect a solution soon.

One such solution is to extend the RSS type to include 'skip points' in the audio file. By 'skip point' I mean a description of which news item starts at what offset. The Podcast descriptor would contain not just the location of the file, but also the contents of the audio file. This would also require a special podcast player, which is able to read and understand the podcast descriptor. Of course, this needs to be standardized so that podcasts from all providers adhere to a single standard. Another advantage of this is that the descriptors could be searched in a standard way and podcast directories are able to show news items and the exact location of those news items in podcast files.

However one problem with this technique is that, it is not easy to make listeners listen to advertisements. It would be easy to skip advertisements if the listener is not interested in it. A second problem is that the accuracy of the podcast descriptor is in the hands of the provider.

Any other solution?