Exalead: Voice to Text

November 3, 2008

A happy quack to the stylish Parisian who alerted me to Exalead’s voice to text demonstration. To use the service, navigate to http://labs.exalead.com or click here. I entered several test queries and looked at the quality of the ASCII. I was impressed. I was able to get useful hits on my trusty query ‘bush and iraq”. My Google queries worked well too. Keep in mind that the system has processed a chunk of audio and video. The voices in the files are converted, indexed, and made searchable. One nifty feature is that if a video contains several references to the query term, an icon on the play bar allowed me to jump from relevant comment to relevant comment. No more serial listening to talking heads. Two happy quacks for the Exalead engineers who worked on this demo. Several other nice touches warrant highlighting:

  1. The system can parse a query such as ‘show me videos about iraq’
  2. Entities are automatically extracted and displayed in a side bar for assisted navigation
  3. A tab allows you to limit your query to audio, video, video on demand, or the entire suite of content.

For me, the most useful feature was the ability to click the ‘text’ link and see the transcribed text of the news show. Here’s a snippet of the machine converted and transcribed text:

the big apple behind the turntable strolling down the house makes tonight in chicago is craig alexander find your way to the bone bloomer whom you’ve gone and only together since the first of the year the brian james van by achieving their goal of crafting and plain old b. s. rock and roll the show tonight is that the hurricane in kansas city that’s a for tonight’s live music on the east coast air midwest for a look at what’s gone down monday night in the south boston that soars southern music reporter john spellman

My recommendation to Exalead is to start processing more content. I would love to have a transcript of the Google lecture series. A collection of security podcasts would be really useful. I don’t like to listen to 50 minutes of lousy audio to find one or two useful chunks of information.

I usually try to remind the French that folks from Kentucky know how to cook chicken correctly. None of coq au vin stuff. We use lard and whatever is growing behind the compost heap. But in this case, I won’t make any reference to cuisine. I will just say, “Voice to text… well done.”

Stephen Arnold, November 3, 2008 from somewhere in Europe

Comments

4 Responses to “Exalead: Voice to Text”

  1. George Everitt on November 3rd, 2008 9:38 am

    I’m not sure what to make of that transcribed text. Are you illustrating the inadequacy of modern speech to text algorithms, even in the face of decades of research?

    “Voice to text … well done”

    Is that irony or cynicism? The text looks like random Bayesian spam filter chaff.

    I’m not trying to be controversial – but… huh? I must be missing something.

  2. Stephen E. Arnold on November 4th, 2008 1:58 am

    George Everitt

    Thanks for your comment. With the addled goose, you have to answer your own excellent questions.

    Stephen Arnold, November 4, 2008

  3. Activeille | Voxalead, un moteur de recherche vocale | Veille et intelligence économique pour les PME TPE on January 30th, 2009 3:14 am

    […] Exalead: Voice to Text (arnoldit.com) […]

  4. Le blog d'Aurigance Développement on February 3rd, 2009 9:25 am

    […] Exalead: Voice to Text (arnoldit.com) […]

  • Archives

  • Recent Posts

  • Meta