Unwired Tech: January 2006

Tuesday, January 31, 2006

Words Are All I've Got

Nearly twenty years ago, when I first moved to the big city of Toronto, Canada, I only knew 3 languages. Court and embassy translators were then in high demand, and I thought I might be able to get some gigs. But despite my interest, I never learned any extra languages until after I moved away from Toronto. It's my guess, however, that like in any other big, busy international city, Toronto translators never thought their job well would run dry. If IBM's "MASTOR" (Multilingual Automatic Speech-to-Speech Translator) prototype technology, which currently runs on Win XP and CE, so on a PDA, ever gets released commercially, translators may have to find new work.

MASTOR is currently capable of taking an English-language audio as input via a microphone, and converting it to spoken Mandarin, Arabic subtitles for a TV broadcast and more. As an armchair linguist myself, I'm doubtful that anyone's current speech conversion technology is perfect or even any where near so. But MASTOR might be more evidence of a coming revolution or discovery regarding a universal translator.

While human translators probably don't have any immediate worries, I might recommend that they either get involved in such speech conversion development, or plan for a career change in their lifetime.

Links/ Sources: PCMag - IBM Strives For Superhuman Speech Tech.

(c) Copyright: 2006-present, Raj Kumar Dash, http://webguru.mathgurusonline.com/tech-watch/

Technorati Tags: tech watch, speech technology, speech-to-text, linguistics

Searching In Audio Podcasts

TMCnet.com and Businesswire both mention a new audio search engine by PodZinger.com. (The Businesswire text is a press release). (Links below).

This is a fascinating idea become reality. Apparently, PodZinger is "based on 30 years of speech recognition research" from BBN Technologies. PodZinger first converts audio podcasts into text, then builds an index of keywords. When a user searches for certain words or phrases, related keywords are presented with visual cues, including "the text surrounding the search term".

As a test, I searched for the specific phrase "U2 concert", got two results, clicked on the second one, and heard a fragment of a podcast with two DJs discussing a U2 concert. I did another test using "poker playing". This returned several results. I clicked on the first result, then waited while the engine skimmed through the audio until about 18 minutes in.

What I found in both cases, though, is that even reasonable quality audio is sometimes not translated to text properly. But that's to be expected, given the multitudes of speech patterns. This technology needs to mature, and the neural network software continually trained. It'll be a while before software can translate audio of a single language perfectly. Let's face it, even humans have trouble understanding each other, let alone the semantics of what's being said.

If you ask me, though, this technology could be a key element in the quest for a universal language translator. The next step after converting English-language audio to text is to convert the text to another language, then to audio in the second language. Bravo, PodZinger and .

Links/Sources: PodZinger, TMCnet - Laura Stotler - PodZinger Search Engine Finds Spoken Keywords in Podcasts, BusinessWire - PodZinger Makes Podcast Searching Fast, Easy and Accurate, BBN Technologies Avoke STX Speech-To-Text Software.

(c) Copyright: 2006-present, Raj Kumar Dash, http://webguru.mathgurusonline.com/tech-watch/

Technorati Tags: audio search, search engine, speech to text, universal translator

Tuesday, January 31, 2006

Words Are All I've Got

Friday, January 27, 2006

Searching In Audio Podcasts