• Visitors can check out the Forum FAQ by clicking this link. You have to register before you can post: click the REGISTER link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below. View our Forum Privacy Policy.
  • Want to receive the latest contracting news and advice straight to your inbox? Sign up to the ContractorUK newsletter here. Every sign up will also be entered into a draw to WIN £100 Amazon vouchers!

BBC having problems with software

Collapse
X
  •  
  • Filter
  • Time
  • Show
Clear All
new posts

    BBC having problems with software

    'Spending Our Cash Recklessly On A Totally Empty System': 6 years and £8.3m later, there's still no sign BBC's foreign monitoring technology Socrates

    Oh look, it's Autonomy...

    The 370-strong BBC Monitoring division, which analyses and translates into English radio and television bulletins in around 100 foreign languages, is having to monitor the world’s media with what one member of the team described as “antiquated and temperamental equipment that belongs in a museum”.

    Socrates was meant to change that. A revolutionary computer system from the Cambridge-company Autonomy, it promised to revitalise the text-based Caversham system with artificial intelligence and the means to translate myriad languages from audio into written English. So sophisticated was this Technology Refresh Project that it was given a lofty acronym based on its anticipated benefits. Specialised, Open source, Collection, Reversioning, Archiving, Tailored, Export, System - Socrates!

    It was supposed to be operational in 2009 but, six years after it was commissioned and despite an expenditure of £8,346,847 (a figure released by the BBC in response to a Freedom of Information request), BBC Monitoring is still using its old text system.
    Behold the warranty -- the bold print giveth and the fine print taketh away.

    #2
    BBC having problems with software

    Urgh, it was always trying to do far too much. People have such unrealistic expectations of Speech Technology. :-/
    Automatic Speech Recognition/Voice-to-Text is complex enough in one language - maintaining 'a myriad' of input languages (and presumably attempting to use Automatic Language Identification to differentiate between them) is bound to not work sufficiently well. Trying to then pair that with machine translation (a field full of its own issues) just takes the p***.

    Bloody hell.

    </the computational linguist's rant>

    Comment


      #3
      I had a fair bit of success with Voice to Text in the late 90s, except for when the phone rang and even with a directional microphone putting a mug of coffee on the desk would generate some random noise. I was fortunately using some software which had been tailored for the British market so I had a chance by speaking RP. I could easily confuse it by breaking into a different accent.

      Even OCR which is a more mature technology is highly dependent on the quality of the source, though it is far better today than when I used it a couple of decades ago. I can still confuse it with the occasional multilingual document.
      Behold the warranty -- the bold print giveth and the fine print taketh away.

      Comment


        #4
        Autonomy? Snake oil salesmen, just as your average HP employee...

        Comment


          #5
          Originally posted by formant View Post
          Urgh, it was always trying to do far too much. People have such unrealistic expectations of Speech Technology. :-/
          Automatic Speech Recognition/Voice-to-Text is complex enough in one language - maintaining 'a myriad' of input languages (and presumably attempting to use Automatic Language Identification to differentiate between them) is bound to not work sufficiently well. Trying to then pair that with machine translation (a field full of its own issues) just takes the p***.

          Bloody hell.

          </the computational linguist's rant>
          Those chaps at Menwith Hill don't seem to have a problem with that kind of stuff but that said, the NSA has a bigger budget.

          Comment


            #6
            It does seem as if they were/are reaching for the Moon, technology wise.

            Even voice message to text conversion, in a limited set of languages, has so far proved impossible to automate entirely, as companies like SpinVox have found out:

            The Voice Message Conversion System (VMCS) worked by combining speech technologies with live learning capabilities and human intelligence. It was developed by the SpinVox Advanced Speech Group based in Cambridge, UK, led by Cambridge academic entrepreneur Dr. Tony Robinson and includes Cambridge University Professor Phil Woodland. The company supported the following languages: English; French, Spanish, German, Italian and Portuguese. Parent companies such as Nuance Communications have claimed that "spinvox is offering something that is impossible to deliver now"
            Work in the public sector? Read the IR35 FAQ here

            Comment


              #7
              Originally posted by SupremeSpod View Post
              Autonomy? Snake oil salesmen, just as your average HP employee...
              They deserved each other.
              Behold the warranty -- the bold print giveth and the fine print taketh away.

              Comment


                #8
                Originally posted by OwlHoot View Post
                It does seem as if they were/are reaching for the Moon, technology wise.

                Even voice message to text conversion, in a limited set of languages, has so far proved impossible to automate entirely, as companies like SpinVox have found out:
                I worked for SpinVox after Nuance bought them. Full automation is possible in this context, but only after training your algorithm with the with agent-transcriptions (and a substantial amount of those). SpinVox's real failure was their PR - the blatant lies about fully automating when they simply weren't ready yet (well, that and the treatment of their offshore transcription centres). They did develop a fairly good automation system alongside their agent transcription and these days a number of the languages offered by Nuance Voicemail-to-Text (formerly SpinVox) are fully automated and deliver adequate results (adequate - not at all brilliant), others work very well with a combined speech rec/transcription system.

                Full ASR however works a lot better on systems where each speaker builds up a profile for himself. Anything ad-hoc (like voicemail messages) can work decently, but will always be much more unreliable. I work on Voice-to-Text/ASR in medical transcription these days, which is an overall much better product and a significantly more sensible application. I work particularly on tailoring and fine-tuning those systems to recognise regional, ethnic, and non-native accents. (I'm a phonetician by training, not a developer in this field). This is the sort of thing that only works where each speaker has a separate account though and isn't easily applicable to ad-hoc ASR.
                Last edited by formant; 7 March 2013, 09:39.

                Comment


                  #9
                  Originally posted by Sysman View Post
                  They deserved each other.
                  It's remarkable how much BS gets sold to gullible ejits in terms of speech technology. Take the Nemesysco Voice Stress analysis stuff for example. Lots of banks and insurance companies bought it (and many are still using it). Just like with your average polygraph, there's simply no science behind it (few good posts/articles here: Nemesysco | Phonetics at Stockholm University Blog).

                  Comment


                    #10
                    Originally posted by SupremeSpod View Post
                    Those chaps at Menwith Hill don't seem to have a problem with that kind of stuff but that said, the NSA has a bigger budget.
                    If they did have a problem they wouldn't go public with it.
                    And what exactly is wrong with an "ad hominem" argument? Dodgy Agent, 16-5-2014

                    Comment

                    Working...
                    X