back to article Fun fact of the day: Voice recognition tech is naturally sexist

Voice recognition systems are sexist: they struggle to deal with female voices compared to male ones. It's a headache that has been lingering for a while in the machine-learning world. The issue was brought back into the spotlight again by Delip Rao, CEO and co-founder of R7 Speech Sciences, a startup using AI to understand …

  1. Long John Brass Silver badge
    Mushroom

    Bullshit and bollocks

    Voice recognition systems are sexist

    It's a narrow training set. Voice recognition systems don't understand a bloody thing I say either as I'm don't have a USAian accent.

    Now it may have problems with understanding male .vs female as well. But to be fair most men can't understand a thing women say either; So it's a bit cheeky to expect a computer to :)

    It's a bird, it's a plane; No its an unholy shit-storm of downvotes :)

    1. Anonymous Coward
      Anonymous Coward

      "No its an unholy shit-storm of downvotes"

      Well I'm just thinking the down arrow points towards your 'brass' and I just have to give it a ding.

    2. teknopaul Silver badge

      Re: Bullshit and bollocks

      A crap sexist joke doesn't stop being a crap sexist joke because you downplay it in advance.

      Try this...

      Before making crap sexists jokes try replacing it with a crap racist joke and see how it stands.

      In this case you probably deserve a few more down votes than you got because tech-bro.

      1. Anonymous Coward
        Anonymous Coward

        Re: Bullshit and bollocks

        > A crap sexist joke doesn't stop being a crap sexist joke because you downplay it in advance.

        Exactly. Tarnishing all men with the same brush is sexist. SOME men can understand women (not me. I am not broken enough yet).

        1. GX5000

          Re: Bullshit and bollocks

          Depends on the individuals doesn't it?

          I listen to these kids speak nowadays and wonder if it's still right to call it English...

      2. Anonymous Coward
        Anonymous Coward

        Re: Bullshit and bollocks

        Oh look another touchy lefty fascist...

    3. Anonymous Coward
      Anonymous Coward

      Re: Bullshit and bollocks

      Voice recognition systems don't understand a bloody thing I say either as I'm don't have a USAian accent.

      There's a lot of truth to this.

      As a Brit in the USA my Siri seems to not understand a lot of things I say.

    4. MrDamage

      Re: Bullshit and bollocks

      > "But to be fair most men can't understand a thing women say either"

      The other day, the missus said "You haven't been listening to a single word I said, have you?"

      I thought that it was an odd way to start a conversation.

      *badoom-tish*

      I'll be here all week. Tip the steak, try the waitress.

  2. Rattus Rattus

    Female voices...

    "...more easily masked by noise."

    Oh how I wish that were true.

    1. Charlie Clark Silver badge

      Re: Female voices...

      Don't worry: cognitive psychological research suggests that men tend to find female voices* more soothing that male ones and, partly as a result of this, difficult to pay attention to for long periods.

      * As long as they're not "frying" their vocal chords or doing Mickey Mouse impressions.

  3. joed

    sex sells

    I'd guess that the choice of Alexa, Cortana and Siri was also dictated by the sexism. I've kept my microphones disabled (equal opportunity for voice recog systems;). Now, I'll be glad if I could also completely disable these irritating voices (agent C in particular). I'll do this in the name of equal rights.

    1. fuzzie

      Re: sex sells

      I've heard different, reasonable-sounding justifications for assistants, and earlier even, voice navigation, using female voices.

      As human we're more acutely tuned to female voices... possibly due to our early, formative years spent around mom. Secondly, noisy environments, e.g. cars, tend to drown out lower frequencies, given higher pitched female voices a slight advantage. And thirdly, people tend to be more compliant accepting instructions from women. The latter could again be a side effect of our early years spent with mothers or possibly a way to bypass the male competitive streak.

      I'd love to hear from anyone with more official knowledge though.

      1. Charlie Clark Silver badge

        Re: sex sells

        And thirdly, people tend to be more compliant accepting instructions from women.

        No, they find female voices less aggressive so their response is differently primed.

        1. Phil O'Sophical Silver badge
          Coat

          Re: sex sells

          No, they find female voices less aggressive so their response is differently primed.

          Maybe they should be aggressive? After I've ignored the satnav for the sixth time, and she is still calmly saying "when possible, make a U-turn", I find it disappointing that she hasn't started to get snarky, or gone off in a huff. Not very realistic...

          1. JakeMS

            Re: sex sells

            "I find it disappointing that she hasn't started to get snarky, or gone off in a huff. Not very realistic"

            Careful what you wish for, it may start saying things like.

            "OMG OMG OMG You missed yet another turn! OMG"

            "Are you listening to me?!"

            "If you don't start listening to me, you got big problems mister!"

            "OMG, you never listen to me, what is wrong with you?!"

            "What did I just say? That's right you don't know BECAUSE YOU NEVER LISTEN!!!!!"

            "Let me out, I don't want to be your satnav anymore! Oh, and I'm taking the radio with me!"

            "Don't ever ask me for directions again! GOODBYE!!"

            *Door slams shut*

          2. Charlie Clark Silver badge

            Re: sex sells

            Maybe they should be aggressive?

            Ahem, you seem to have missed the point: man hears another man and hears a potential threat This is hardwired into our brains so difficult to mimic. One of the reasons why some women develop deeper voices when they work with a lot of men; Maggie was apparently coached for this.

            "Frying" is the partially the result of untrained attempts of women to lower the pitch of their voices.

      2. Anonymous Coward
        Anonymous Coward

        Re: sex sells

        The choice of sex for a synthesised or recorded voice tends to depend on the country. This was discussed at length when the Austin Maestro came out in 1988, and again, repeatedly, every time a new internationally marketed product came out with a synthesised or recorded voice. I wonder if anyone has made a table to summarise this, along the lines of UK: female, Germany: male, ...

      3. Carpet Deal 'em Bronze badge

        Re: sex sells

        "I've heard different, reasonable-sounding justifications for assistants, and earlier even, voice navigation, using female voices."

        One of my local grocery stores has the typical female voice in English mode on the self checkouts, but a male voice in Spanish mode(US). Make of that what you will.

  4. Anonymous Coward
    Anonymous Coward

    females generally have higher pitched voices

    "They also tend to be quieter..." What?

    "it's more easily masked by noise, like a fan or traffic ..." What?

    Computers are relatively unbiased in raw data. A 1 is a 1, a 0 is a 0. This means human perception on loudness/ quietness shouldn't even be in here, as the staff can simply increase the sound volume as necessary for the AI. Also if the vocal range exists and no other frequency is directly overlapping and interrupting it, you will get the human vocal from the mix of sounds regardless of fan or traffic. In fact, it takes a free open-source audio software to quickly manually single out the human vocal range in a music file, easily without AI. So saying female voice is quieter or masked voice by fan or traffic not a good excuse for the lack of training on the AI with female voices.

    1. frank ly

      Re: females generally have higher pitched voices

      "So saying female voice is quieter or masked voice by fan or traffic not a good excuse for the lack of training on the AI with female voices."

      It's not a good excuse for lack of training with female voices, as opposed to male voices, which is one problem with the training on data sets. However, if external noise due to traffic and domestic/office noise is in the same frequency range as the female voice, or if the female voice is generally quieter, then the S/N ratio will be lower at the recorded source.

      In this situation, you can't improve things by turning the volume up. If you ran some fancy noise reduction process over it then the male voice recordings would be improved as well and so the male recordings would still have a better S/N ratio than the female recordings.

  5. EveryTime

    I expect that attitude has a little to do with it.

    My wife sharply orders Alexa do something, then yells angrily at it when it does not understand.

    I'll repeat the same request, much more quietly, in a measured tone. That almost always works, and the success makes both Alexa and me quite unpopular.

    1. Anonymous Coward
      Anonymous Coward

      My wife sharply orders Alexa do something, then yells angrily at it when it does not understand.

      Half of my text messages contain "Eff you Siri!"

  6. Paul Hovnanian Silver badge

    My wife ...

    ... says I don't listen to her.

    I thought that was a funny way to start a conversation.

  7. Rebel Science

    Sexist you say?

    It's time to ban them on Twitter, FaceBook, Instagram and YouTube for hate speech. Also, detain them at the London Heathrow airport for 3 days and confiscate their passports. But do let the jihadis back in, by all means. LOL

  8. Tromos

    Quieter???

    Whoever came up with the idea that female voices are quieter has obviously never been on a bus with a dozen or so schoolgirls.

    1. Charlie Clark Silver badge

      Re: Quieter???

      Volume != pitch

    2. Teiwaz Silver badge

      Re: Quieter???

      obviously never been on a bus with a dozen or so schoolgirls.

      Why merely a bus? Amplification effects?

      I'd have thought anywhere with a 'dozen or so schoolgirls' would be enough to cause lasting auditory and psychological harm*

      * Often it's not just how loud they are talking, it's whether you can hear what they are saying...

      1. Mage Silver badge

        Re: whether you can hear what they are saying...

        WHY would you WANT to know what a bus load of school kids (of any sex) are saying? Their own mothers and teachers would probably rather not know.

    3. Mage Silver badge

      Re: Quieter???

      I was on a flight from Austria to UK. A large school trip of returning teen school girls. At one stage the Captain came on intercom and suggested if they were not quieter he'd get permission for the cabin crew to put them off.

      It was dreadful.

      However a bus load of teen boys to & from a rugger match is worse.

      1. Anonymous Coward
        Anonymous Coward

        Re: Quieter???

        At one stage the Captain came on intercom and suggested if they were not quieter he'd get permission for the cabin crew to put them off.

        He could at least wait till they land somewhere.

        1. Mage Silver badge

          Re: wait till they land somewhere

          I'm not sure that landing was suggested. There was a shocked hush for a few minutes and it didn't get as loud again.

          Obviously he'd not have the crew attempt to open a hatch while at altitude.

    4. Anonymous Coward
      Anonymous Coward

      Re: Quieter???

      " has obviously never been on a bus with a dozen or so schoolgirls."

      Should you be hanging out on school buses at your age ... ?

  9. Anonymous Coward
    Anonymous Coward

    Of course,

    … it's due to the patriarchal technocracy: if fans had drowned out their words, the male engineers would have made bloody sure the things were less noisy from the get-go.

    (Interesting fact: you can't roll your eyes ccw while rotating your wrists cw.)

    1. Flocke Kroes Silver badge

      Re: Eyes widdershins, hands turnwise

      You can, but it takes a little practice. Same with drawing a circle with one hand and a triangle with the other or rubbing your tummy with a circular motion while patting your head. Learning any of these important skills makes learning the next easier. When you are good at a few of them you will be able to do different (simple) tasks with each hand. I use it for cooking and convincing guests who see me cook that I am a bit strange - or at least slightly stranger than they already knew.

      1. Anonymous Coward
        Anonymous Coward

        Re: Eyes widdershins, hands turnwise

        Same with drawing a circle with one hand and a triangle with the other or rubbing your tummy with a circular motion while patting your head.

        I misread that, and had to stop to count my number of arms.

    2. Steve Davies 3 Silver badge

      Re: Of course,

      Don't worry fellas...

      http://www.bbc.co.uk/news/av/business-43389704/a-co-working-space-only-for-women

      1. TRT Silver badge

        Re: Of course,

        That co-working space is an abomination. Not only is it inherently gender bias in the workplace, it also screams "oh, must be protected, can't compete on an even playing field"; which is utter bollocks.

        The only "women-only" zones I consider acceptable, barring areas of undress, are places like domestic abuse shelters, and even THEN that's not barring access to an abusive partner in a female-female relationship.

  10. Mark 85 Silver badge

    Something is rotten in AI....

    So what is the underlying reason? Males doing the programming and not women? Lack of a proper data set? Or is it something else like women's voices change (volume, pitch, even inflection of words)more then men's during stress, etc. and the computer can't cope?

  11. Dodgy Geezer Silver badge

    ....It’s an inherent technical problem down to the fact that females generally have higher pitched voices. They also tend to be quieter and sound more “breathy”....

    Only in colonial, Empire, 'old white males' science. In the new wonderful post-modern world there is no problem, because all voices are equal in every way, and everything they say is of equal merit and validity.*

    * Except white male voices, of course. And any traditional working class person from the North of Watford.....

    1. Dodgy Geezer Silver badge

      One down-vote? I didn't know there were Guardian readers on The Register.... !!

  12. Destroy All Monsters Silver badge
    Windows

    Diversity Victimization Bandwagon Leaving Platform 2

    ...while in the real world, we just have crap products.

    Don't use or fix.

    FFS.

  13. Anonymous Coward
    Anonymous Coward

    Are you kidding me?

    "females generally ... tend to be quieter "

  14. DavCrav Silver badge

    "It’s an inherent technical problem down to the fact that females generally have higher pitched voices. They also tend to be quieter and sound more “breathy”."

    So in other words, women can't speak as clearly as men, on average. It's like saying facial recognition programs have trouble with black faces because they scatter less light, so in training that's fine because there's a big spotlight, but in real life it doesn't work as well, and can never work as well.

    This beautifully complements an article in The Guardian the other day that stated that algorithms favour white men because there aren't enough female coders, along with a lot more idiotic garbage.

    1. Charlie Clark Silver badge

      facial recognition programs have trouble with black faces because they scatter less light

      This isn't true, they just scatter it differently. Some other commentard was able to provide a bit more information on how this is dealt with in the photographic industry.

      But the general point stands that there may well be technical factors that cannot be effectively compensated for. This leads to opportunities for those wishing to seize them: either in the market place or on the PC gravy train.

      1. DavCrav Silver badge

        "This isn't true, they just scatter it differently. Some other commentard was able to provide a bit more information on how this is dealt with in the photographic industry."

        I would have thought that black surfaces reflect less light than white ones. In particular, in low light levels, it's harder to spot black people in real life than white people.

    2. Mage Silver badge

      an inherent technical problem

      Like maybe AI is PR, everything is hyped. Deep Learning & Neural networks are also totally misleading terms.

      It's software. How often does a program not have crap bits and no-one can figure why?

      Sometimes like clippy, ribbon or Metro/ModernUI we do know why but whoever is in charge isn't listening.

  15. Adam 52 Silver badge

    "Voice recognition systems are sexist: they struggle to deal with female voices compared to male ones."

    Others have alluded to it, but I'll just highlight this paragraph. Struggling to understand is not sexist. There is no prejudice or discrimination on the basis of gender. There is a difference in behaviour, but that's not the same thing.

    All of which the article goes on the explain, so the only justification for that first paragraph is a "journalist" wanting to "sex up" the story or mislead the reader.

  16. Anonymous Coward
    Thumb Up

    Scientist says women all sound the same

    No, not something I would assert as true, but a stab at the kind of headline many journos use to manufacture controversy out of science. And she happens to be a female scientist[1] talking what looks like good commonsense[2], unencumbered by the need for Political Correctness and fear of transgression.

    Shame then that a Reg journalist stoops to introducing "lack of diverse training examples" in reporting it. The Inquisition is always with us.

    [1] I presume, from the name Rachael.

    [2] That is to say, it's intuitively credible, and doesn't look like Agenda Science.

  17. Primus Secundus Tertius

    Telephonists

    I am surprised by this report.

    Telephone companies traditionally stated that they employed women telephone operators because their voices were clearer on the line.

    But perhaps the real reason was that women's wages were lower.

    1. FrankAlphaXII

      Re: Telephonists

      If it was AT&T saying that, you can fucking guarantee it was because wages were lower and they knew CWA wouldn't raise a big fuss about it during CBA negotiations.

    2. Nick Ryan Silver badge

      Re: Telephonists

      Telephones, i.e. landlines, have a very limited frequency range therefore there may be some truth to this.

      1. Mage Silver badge

        Re: Telephonists

        Also allegedly a reason for female radio operators in WWII, though it might have been anxious staff procurement wanting to avoid having female appointments blocked.

        However I don't think it's a frequency response issue, but the idea that fainter female voice with more background noise is easier to understand. So I suspect that the software is poor rather than anything inherent about female voices, which can be clearer (certain parts of Brooklyn, Dublin, Yorkshire and West Country may be exceptions).

        I expect they have poor choice of samples, despite claims that it's some other problem.

        The only difference I see from 25 years ago is that it seems to skip the personalised user training and give about the same performance, poor. The phone menus with voice recognition are ghastly compared to numbered menus, which are ghastly. None of it is to make a better customer experience but to save money.

        Same with evil automated checkouts, which encourage some people to cheat. No doubt the thefts are less than cost of a checkout operator. I don't expect transparent & honest figures for how accurate Amazon's invisible checkout is. Next step is to ban customers from aisles and have Argos style terminal (identical to online website / phone app) and you queue to wait for shopping basket to appear. Pay by paypal, IBAN, plastic or RFID phone.

    3. earl grey Silver badge
      Facepalm

      Re: Telephonists

      My understanding was that the original operators were men but that when WW I started the men were all shuffled off to war and the jobs (and then lower pay) went to women. It stayed with women; as did a lot of factory jobs through WW II until the war was over and then the underpaid women were replaced by higher paid men doing the exact same jobs.

    4. Nick Kew

      Re: Telephonists

      Telephone companies traditionally stated that they employed women telephone operators because their voices were clearer on the line.

      That's entirely compatible with the report.

      If female *voices* sound more alike than male ones, that would tend to make their *words* easier to follow in adverse circumstances like a noisy/low-bandwidth line.

    5. Paul Hovnanian Silver badge

      Re: Telephonists

      Police dispatchers as well. Mainly women, as they tend to be easier to understand over low bandwidth channels in noisy conditions.

  18. Anonymous Coward
    Anonymous Coward

    Female voices and noise

    Recently went to a 80th birthday party of a lady I know. Most people there were women, but with a room full of them talking at full speed and volume it just turned into noise, while the lower pitch (and slower pace) of men talking came through quite well.

    I suspect that if women tried the Margaret Thatcher approach of lowering their pitch and slowing down then voice recognition would work well.

  19. Anonymous Coward
    Anonymous Coward

    Click-bait for misogynists.

    Stop it.

  20. FrankAlphaXII

    IVRs suck, though I know of a few that have interesting behaviors. There's one that a bank uses which will automatically get you a human if you say Fuck or bullshit to it. I'm sure they have plenty of recordings of me saying "fucking bullshit" just to get a rep.

    Yes, IVRs suck. They make some things that should take 10 seconds take a few minutes. However, trying to push a sexism angle because IVR's are god awful is pure Bay Area intersectional stupidity at it's finest. What's next? Soft keyboards are ageist or ableist because they require a good amount of dexterity to use?

  21. tiggity Silver badge

    " I feel frustrated most ASR systems fail terribly with Indian accents"

    Odd, as with so much IT stuff outsourced to India it should not be hard to get plenty of training data...

    Or, they could just ring about any customer service helpline in the UK (or wait to get "you have a problem with your computer" scam phone calls)

    I remember many years ago (before ubiquity of Eastenders had increased London accent exposure to non Londoners) having to translate between a Cockney and a North Scotland person, both had quite strong "local" accents. They could not understand each other, I was from a location between the 2 & could understand both accents so conversations were conducted via me (they could also, fortunately, both understand my accent). So if humans struggle to understand "native speakers" a few hundred miles different to them, no surprise AI has issues

  22. DCFusor Silver badge

    No politically correct way to say some things.

    Could it be that people speaking (for example) English as a second language simply don't speak it as well, regardless of accent? I'd think they miss some of the crucial distinctions and oddities that make it possible for us to separate near-homonyms from a native speaker. Further, it's well known (to speech researchers anyway) that most people don't actually "say what they say" - syllables run together and so on - this was a major problem initially in speech recog (which was a field I did a lot of time in). Humans, unlike computers (at least then) make _extensive_ use of context to overcome this and related issues when "understanding" another's speech (which is far more to the point than mere recognizing, when even a perfect transcript would have those missing or run-together syllables).

    The "not pronouncing what you thought you said" issue was a huge eye-opener for many workers. It's easy to perceive it now due to non linear audio editing tools that let you examine small excerpts, though.

    Further, when the speaker is "thinking" in their base language, they use grammatical constructs native to that language but not English, which tends to defeat our human predictive understanding, or "sounds to meaning conversion".

    I know that even I have some issues with strong accents, and also that even long time exposure to same doesn't overcome them. I'm no longer in the field, but I'd bet an analysis along these lines would bear out these observations.

  23. earl grey Silver badge
    Unhappy

    She hate me

    The voice system in my neighbor's truck won't listen to me at all. I generally don't mumble and try to enunciate well using plain (American) english. it is nearly impossible to get that crappy system to listen and understand me; you would think she was my wife or something.

  24. Anonymous Coward
    Anonymous Coward

    So AI has trouble understanding women?

    That‘s final proof, I must be an AI.

    Anon because you know.

  25. Rocketist
    Holmes

    High pitch, accents, background noise -

    If all of those make it difficult for ASR systems to do their job, it might be they‘re no better nor worse than mere humans. I sometimes find it difficult to understand people talking eith an accent - remember a London cabbie saying what sounded like „o‘rite mite“ when apparently they meant „alright mate“. It can be hard tounderstsnd, no doubt.

    Obviously with the accents a possible solution will be to train the ASR with all sorts of them, and possibly add some better heuristics for short phrases. With voice pitch, a higher frequency should carry fewer overtones in the audible spectrum, so maybe the signal quality really is worse. The test case for a fair comparison would be high-pitched male voices. Say boys. And f it‘s so, the AI will simply have to increase its effort with the pitch of the voice. As we do.

  26. Anonymous Coward
    Devil

    And so it 'should' be

    Its owner/for Women, it should be sexist (Biased towards women) and if it's owner is a man/for men it should also be sexist, biased towards men.

    Men are naturally attuned in certain ways toward women and vice-versa due the the inherent biological needs of both and humans to procreate.

    That said, society in Australia is supposed to treat all as equals and that's fine by me. but people come from their own perspective. A little banter is OK, too much and it is too much!

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Biting the hand that feeds IT © 1998–2019