Hi-Tech

Microsoft's speech recognition system reaches human-level accuracy

Microsoft's speech recognition system reaches human-level accuracy

Almost a year ago Microsoft revealed that it had bested IBM by hitting a word error rate (WER) of 6.3% (in the industry-standard Switchboard speech recognition benchmark), dropping that to 5.9% just a month later - and at the weekend the firm announced that it's now achieved a WER of 5.1%. This was generally considered to be the average human error rate, but further work by other researchers suggested that 5.1 percent was closer to the mark for humans professionally transcribing speech heard in a conversation. This was consistent with prior research that showed that humans achieve higher levels of agreement on the precise words spoken as they expend more care and effort.

Following the success of Microsoft's speech and dialog research group in 2016, it was put to them to get its word error rate down from 5.9 per cent to 5.1 per cent to match that of those involved in a multi-transcriber process and it done just that. But now the speech recognition tech can look at context for clues. This is done by tasking either humans or a machine to transcribe recorded telephone conversations between strangers on topics including politics and sport. As the technology better emulates human speech, fewer user errors will take place.




According to Microsoft, the company's researchers were able to reduce the error rate by about 12 percent by using a series of improvements to neural net-based acoustic and language models. Additionally, our approach to combine predictions from multiple acoustic models now does so at both the frame/senone and word levels. "Microsoft's willingness to invest in long-term research is now paying dividends for our customers in products and services such as Cortana, Presentation Translator, and Microsoft Cognitive Services", said Xuedong Huang, technical fellow at Microsoft, in a blog post. In simple terms, this means the new language model allows the system to use the entire preceding conversation as history when trying to clearly determine specific phrases.

Much of the work is powered by Azure GPUs.


  • There are Barcelona directors that shouldn't be there, blasts Neymar

    There are Barcelona directors that shouldn't be there, blasts Neymar

    It comes two days after Neymar criticised Barcelona's under-fire board after scoring twice in his home debut for PSG. Barca also settled a case with the Spanish tax authorities over the transfer for 5.5 million euros previous year .
    Here's what Nintendo's new $80 mini Super Nintendo looks like in action

    Here's what Nintendo's new $80 mini Super Nintendo looks like in action

    The SNES Mini releases on 29 September, though as we've mentioned, getting a pre-order is going to be tough at this stage. SNES-era games wore their arcade influences on their sleeves, which means they nearly revel in killing players.
    Expect Crude oil to trade positive: Sushil Finance

    Expect Crude oil to trade positive: Sushil Finance

    Drillers lost five rigs in the week to 18 August, the biggest fall since January, bringing the total count down to 763. Analysts said data signaled the oil market was starting to rebalance and supported oil prices last week.
  • Gov. Doug Ducey to skip Donald Trump rally in Phoenix

    Gov. Doug Ducey to skip Donald Trump rally in Phoenix

    The biggest outcome of Trump's unorthodox governing style may be seen in Flake's re-election effort. "He's toxic!" Trump wrote. In case you've been living under a rock, we'll say it one more time: The president is coming to Phoenix Tuesday.
    Led strikes to isolate IS in Raqqa kill over 170

    Led strikes to isolate IS in Raqqa kill over 170

    The activist-run group Raqqa is Being Slaughtered Silently said 32 people were killed in airstrikes on one neighbourhood alone.
  • Hundreds Show Up at Broadmoor Library to View Eclipse

    Hundreds Show Up at Broadmoor Library to View Eclipse

    Perhaps 100 of those at the library were also disappointed, as more than 300 pairs of glasses were depleted in about a half hour. Asked why he, his family, and friends, came to Cleveland to view they eclipse, he was very candid.
    Public finances record July surplus but overall borrowing up

    Public finances record July surplus but overall borrowing up

    This reflected new methodology introduced in the current financial year that smooths out corporation tax revenues over the year. A Treasury spokesperson said: "We are making good progress in strengthening our public finances and living within our means".
    Global  cricket returns to Pakistan with WI, SL tours

    Global cricket returns to Pakistan with WI, SL tours

    Now that worldwide cricket is set to make a comeback in Pakistan, former Pakistani pacer, Wasim Akram sent an emotional message. Cricket matches have not been played in Pakistan since the Sri Lankan team bus was targeted by unidentified assailants.
  • Bixby Voice now available in 200 countries

    Bixby Voice now available in 200 countries

    Samsung announced that the voice capabilities are now available in more than 200 countries and territories across the world. Not only smartphones, but we're talking about vehicle manufacturers, and others like audio makers are coming to us [too]".
    Chilean court approves limited abortion for first time in almost  30 years

    Chilean court approves limited abortion for first time in almost 30 years

    Correction: August 21, 2017 12:00 am - A previous version of this story misspelled Michelle Bachelet's first name as Michelet. Pinochet's law had made Chile one of just a handful of countries worldwide to outlaw the procedure without any exceptions.
    Nationwide bank strike today

    Nationwide bank strike today

    According to union leaders, around 10 lakh bankers working in around 132,000 branches will be on strike on Tuesday. With branches closed, automatic teller machines (ATMs) might get empty sooner than normal .