Desktop

Ask something
Discussion Locked

Voice Recognition-Not Perfect and Needs Regulations

0 helpful
10 replies

by bjc_199

Closed to new repliesThis discussion is closed. It stays readable, but it takes no new replies.

For a while I had worked at a local Community college in Haverhill, Massachusetts, near where I live in Methuen, Massachusetts as an Assistive Technology Assistant. As part of my job, I worked with a voice recognition software called Dragon Dictate/Naturally Speaking. The company who vended it out to us claimed that it would accurately recognize a person's voice after 45 minutes of training. The truth of the matter was that it took a good committment of training to even get the software to recognize one's voice accurately, if at all. Even after all of the trainign that was put into it, it still had accuracy flaws in recognizing people's voices.

Today, lots of companies like Sony and 800 directory use voice recognition technology as their primary automated telephone answering services. 800 directory uses a program known as Tellme and Sony uses a program known as Max. One of the programs does not recognize numbers too well. For instance, when I called Sony to seek hekp for a PDA that I owned, it asked me the modle number, which I spoke through their automated system called Max. The model number I was trying to get Max to understand was PEG-SJ-22. Max kept on saying, no matter how I put my numbers "Did you mean PEG-SJ-20. I kept saying no and it finally kept saying, "Let's see if I can help." Finally after all of that frustration I got through to someone and had complained about it on several occassions and to my knowledge they did nothing to fix Max's voice recognition accuracy. Another example is 800 directory's Tellme software. I asked it to look up a number and it from time to time did not understand what I had to say and it kept replying that we did not find any match. On another occasion, I used voice recognition with another company that deals with magazines. I kept saying the numbers 400 or 4 0 0 and it kept repeating did you mean 100. Becoming frustrated I handed someone else the phone and they could not even get it to recognize the numbers 4 0 0 and told the customer services agent that their system is not recognizing things too well.

Voice recognition programs are good for people who are physically challenged and need to use it because they either do not have use of their hands or good eyesight to push the keypads on the telephone. But to have it as a primary automated telephone answering service without no requirement of having a contingent alternative automated system or better yet, a live person come over the phone is just bad business and bad policy. Especially with directory assistance. Even more troubling is when there is only an operator there from 9:00 AM-9:00 PM Mondays-Fridays and on the off hours, only a voice recognition system that is not perfected to a 99% degree of accuracy or better with no contingent system in place when the business closes, and even more so on weekends when you someitmes need a number is also bad business and bad policy. Both, the FCC, the FTC and our legislators needs to step up to the plate with this voice recognition technology and insist that businesses who are going to use voice recognition technology as their primary automated system must also have a live person there at all times or an alternative way of communicating with an automated system and without passing the costs onto the consumers.

10 replies

by oz_media

These problems have been resolved in the recent PBX wars. Again it was up to the phone companies to fix an issue that software coders never could.

IOt is REAllY expensive technology but new INteractive Voice and Automated atendant products by the top manufacturers are turning out realy flawless recognition software.

My network in my New Westminster office allows me to voice in to the utoi attendant then anyone or any extension with a cold, drunk, tired, slurred and I've even tried fooling it but it always gets the answers right and gets me to the right people, voice mail, email coverted to MP3 etc. This technology is used by many of the larger local companies, the Parliment Buldings and several city halls on the island for customer service.

That also coupled with true voice over IP telephony, not just network based switching.

NEAX 2000IPS, Nortel BCM, NEAX IVS.

by network_driven

Many of the issues around voice reco programs, is in the learning curve of the user. Some programs are looking for an exact match to the responses recorded in the program. Most programs cannot filter background noises, and I have walked through a program and on to an operator with out saying a word. If I have a low battery or add verbage to my answer thats not in the code will send the program out for a closest match. A soft voice seems to fly with less flaws, than a man with a deep voice. Some times the pitch is beond the reco's capabilities. Speech is improving daily from consumer demands. Not by government intervention.. Regards

by oz_media

They can save a tonne of money by removing live staff from these positions but do they invest it in thier telephone system? NOOOOOOOOOOOooo!

If I catch a company with such a problem, I get out my best court suit and go knock the door.
selling them solutions instead of waiting for them to improve not only fixed the problem but it has paid me many times.

Put together an alternative and get the sales rep to spiff you for the lead.

by djent

Voice recognition is an evolving and imperfect technology! It is subject to a multitude of variables, regional dialects, voice frequency, background noise, mic sensitivity, frequency clipping and last but not least software. Some of these problems may never be solved to your satisfaction due to the physical limitations others may improve. Some of the problems with Dragon can be improved with a good quality mic.

by oz_media

You said: "It is subject to a multitude of variables, regional dialects, voice frequency, background noise, mic sensitivity, frequency clipping and last but not least software"

Now this is something I have ONLY seen in software based voice recognition. Not anyone elses systems though.

by djent

I missed your point. All VR is software based, locally installed or IVR used at a phone number called. My response to the original post was that it is probably not possible to get 99% accuracy and mandating this by regulation would place every VR vendor in violation.

by oz_media

I was referring to the new pBX based and older IVR systems as being hardware as opposed to the Dragon etc. That is simply software.

I've have never fooled an NEC NEAXIVS2000 even if I purposely slur, mumble or change my pitch. It is a great selling point when demonstrating the system. Customer can't foot it, I can't etc.

That's about as good as your going to get for a while I'm sure. I've NEVER had a problem with a decent system, but cost needs justification, you're not looking at a wee $5000.00 install.

Even a live person will make more mistaked comprehending speech and accents than my IVR does.

This I see as a hardware solution, this is not cheapo IVR by any means but a dual processor PBX VoIP and UM switch.

I know exactly what you are al referring to though, as soon as I hit a cheapo system, it is noted as a sales target. With enough people complaining and bypassing the IVR to access a live operator, it doesn't take much more than some voice testing to convince them they bought the wrong product and saving money wasn't the way to go.

by network_driven

I'm glad to hear that Oz Media has a better pbx system than the one my co. has. I do not always get the person I ask for :( . However this frustration is normal. We are going through a
learning curve and with all of the money that the venture capitalist are spending. I can honestly "say" its here to stay.
Don't worry about the Regulations though. I'm sure our government will be writing up a tax plan that will........ Well ! its going to be big...

by oz_media

I do actually have a full NEAX 2000 IVS PBX at home. True VoIP, UM etc. It's is my line of work sorta, a worder for a phone company and now remote admin for them.

I have an off site PBX that thiers is backed up to and I can moanage VoIP bandwith for them and thier clients. They are also a business line provider for the local telco. It's not like I ran out and spent a few hundred thousand for a telephone system or something. I will brag happily that my switch is more reliable and has more features than the local Telus CO provides. So in all honesty, I wouldn't be surprised if it does have better VR than your local CO.

by hipaa guy

I work for a large Ophthalmology practice that is quite technically advanced. We have been using digital voice recorders lately and have tried some of the offshore transcription services but have found the error rates are too high for our tastes. We want to evaluate voice recognition software for our transcriptionist. Has anyone used the Dragon Naturally Speaking Medical version? Or any other products?

This discussion is closed to new replies.