voice (phone) to data (database) where to look

[moav]

This really isn't a mac or pc question... but I am trying to figure out how to convert phone voicemail messages into text. Why you ask?? We'll this is why I need something to do this unless you guys have a better idea. Our business recieves about 1000 phone messages a day requesting free literature(brochures) At the present time we have 5 old fashion anwering machines(90 minute cassettes in each) that record between a 20 to 30 second message (name, address, city, etc..) We then have 5 gals listen to these tapes and then input these addresses into a database that will check against dupes and such and then shot into our pitney bows bulk mailer. I would like to somehow get these 5 gals who cost me about 100k a year down to 1. So, I need a system that can take voice messages and convert them into text so that we can then enter them into the database for chekcing purposes. If anyone could whip me up a quick app that can do this I would appreciate it ; )

I would love to hear any ideas on how I can acomplish this??

 

[edX]

what, you want us to help you fire 4 people? i don't think i could sleep at nite if i gave you that answer under those conditions.

 

[binaryDigit]

You should look around and see what type of voice processing packages are out there. You don't mention your budget, but I'm sure if it's high enough, you'll find a solution

If it's low, then you can try things like ViaVoice, combined with some IBM voice to xml stuff, combined with some post office address cleaning software. Note that this is still a sketchy solution at best. There are many issues involved in automating something like this.

If you allow a voice solution that is fully automated, you most likely have to have the users spell things out vs just saying them. This means that each call takes longer, which means that you might have to expand the number of lines, which means more money.

If you have an automated system, the error rate is bound to be high, your goal of one person is probably not realistic since you'll have to have people go over suspected errors (e.g. addresses that don't verify). The act of error checking is usually slower than simply pumping in data, which means that the number of messages they process in this fashion will be lower than just entering data.

What is your customer base like? If it's predominantly native english speaking males, then life is good. However if you have a high percentage of non native English speakers (or any speaker with a strong accent), or females or children, then you are going to get higher error rates. Having lots of foreigners also makes it more difficult to do things like implement dictionaries because of uncommon names/spellings.

Anyway, take a look out there, I'm sure there are plenty of others with the same issues. If you try to roll your own, at least you have the advantage that your service is non realtime and you can try different solutions with too much impact on your primary business. It's been a while since I've looked into voice processing so the state of the art may have advanced in the last few years (telcos and telecoms have large budgets for this type of thing, so check out companies like Lucent/Avaya/AT&T/SWBell etc.

Please do reply with any interesting information you find though. And if you still need someone to roll something for you, let me know

 

[moav]

I do appreciate the help. In 22 years we have never had a layoff...I'm sure we can find places for our gals ; ) They hate doing it since it is such a tedious task. Yeh, we are trying to keep our budget on this thing somewhere under a grand a month. I will start calling around. Yes, unfortunetly 99% of our callers our women and of those probably 1/4 have strong southern accents. We were thinking of doing the IBM via voice thing at least that way it's digital to digital. We actually tried a few years back with Dragon Naturally speaking just pluging the tape into the mic port and got about 66% of the information correct. The problem is that these kind of voice recognition is based on one users voice and training is involved. I didn't even think about the spelling the words out. I believe we can get them to key in their phone #'s and zip codes with the keypad so all they would have to speak would be their names and address.

Thanks again