http://thedavisblog.com/blog/?p=691

</sarcasm>

The total  potential population calling into  app is only 60,000 calls during
the 4 week period; control would receive 30K and experimental would receive 30K
both evenly distributed so as to avoid any variables associated with
demographics.

--- In vuids@yahoogroups.com, Todd Chapin <toddwchapin@...> wrote:
>
> Hi Mark,
>
> That will depend on a few factors.
>
> - Your confidence level
> - confidence interval
> - the total population
>
> If we assume the desired confidence level is 95%, that a +-5% confidence
> interval would be sufficient, and that the total potential population
> calling into your app is 1000000, the expected sample size that you'd need
> so that any differences in behavior detected have a chance of being
> statistically significant would be 383.
>
> If you wanted to be more stringent, and set your numbers at 99%, 1% with the
> same population, the sample size would be 14267.
>
> See
>
> http://www.surveysystem.com/sscalc.htm
> and
> http://www.macorr.com/ss_calculator.htm
>
> for sample size calculators and explanation of the concepts.
>
> Todd
>
> On Wed, Nov 11, 2009 at 9:22 AM, Mark <webb.m@...> wrote:
>
> >
> >
> > If I am setting up a control UI vs an experimental UI to measure results,
> > how many calls represent an acceptable volume to analyze results
> > (results=containment) and make a decision? Would 30,000 calls per server set
> > (sets = control vs experimental) over a one month period be adequate? So
> > control would get 30K and experimental would get 30K over the same month and
> > same days, with same random mix of calls from demographics (same mix of area
> > codes and MSA's etc). Control= existing IVR UI. Experimental = new UI.
> >
> >
> >
>
>
>
> --
> !Todd erases a red dot!
>

That's what I thought. I just wanted to get the group consensus.

Do I calculate it on the total calls presented (60K) or the total calls sent to
each group (30K to control and 30K to experimental)? Looks like you've
calculated it based on the 30K per group not the total 60K.

I am rounding, but it's 1/173 using 30K calls? That's 0.0058 (again rounding). A
half of a a percent.

--- In vuids@yahoogroups.com, Peter Leppik <pleppik@...> wrote:
>
> The question you're asking basically boils down to what's the margin
> of sampling error.
>
> The rule of thumb for sampling error is 1/sqrt(n) where n is the
> number of trials (in this case calls).  With 30,000 calls, your
> sampling error will be about 0.58 percentage points--so if the
> difference is containment is more than this, you can be fairly
> confident that it's statistically meaningful.
>
>        -Peter
>
> On Nov 11, 2009, at 8:22 AM, Mark wrote:
>
> > If I am setting up a control UI vs an experimental UI to measure
> > results, how many calls represent an acceptable volume to analyze
> > results (results=containment) and make a decision? Would 30,000
> > calls per server set (sets = control vs experimental) over a one
> > month period be adequate? So control would get 30K and experimental
> > would get 30K over the same month and same days, with same random
> > mix of calls from demographics (same mix of area codes and MSA's
> > etc). Control= existing IVR UI. Experimental = new UI.
> >
> >
>
> _____
> Peter U. Leppik
> CEO
> Vocal Laboratories Inc.
> pleppik@...
>

Hi Mark,

Do you think you should take in account call-cases categories?

maybe I'm out of the subject, cause I talk about my experience of SLM training
not of usability experiment, but with SLM, to determin the amount of data needed
to have a statistically valid model, you usually look at the less-frequent call
case, and make sure that case gets enough chances to be well represented. For
instance you have a portal which distributes calls between 20 main categories
and the less frequent category has an estimated 1% frequency, if you want
minimum 1000 occurences per category, then you need 100*1000 calls...

best regards
---
Ariane Nabeth
tech director Naturalvoice
HOTLINE
www . hotline . fr
--- In vuids@yahoogroups.com, "Mark" <webb.m@...> wrote:
>
> That's what I thought. I just wanted to get the group consensus.
>
> Do I calculate it on the total calls presented (60K) or the total calls sent
to each group (30K to control and 30K to experimental)? Looks like you've
calculated it based on the 30K per group not the total 60K.
>
> I am rounding, but it's 1/173 using 30K calls? That's 0.0058 (again rounding).
A half of a a percent.
>
> --- In vuids@yahoogroups.com, Peter Leppik <pleppik@> wrote:
> >
> > The question you're asking basically boils down to what's the margin
> > of sampling error.
> >
> > The rule of thumb for sampling error is 1/sqrt(n) where n is the
> > number of trials (in this case calls).  With 30,000 calls, your
> > sampling error will be about 0.58 percentage points--so if the
> > difference is containment is more than this, you can be fairly
> > confident that it's statistically meaningful.
> >
> >        -Peter
> >
> > On Nov 11, 2009, at 8:22 AM, Mark wrote:
> >
> > > If I am setting up a control UI vs an experimental UI to measure
> > > results, how many calls represent an acceptable volume to analyze
> > > results (results=containment) and make a decision? Would 30,000
> > > calls per server set (sets = control vs experimental) over a one
> > > month period be adequate? So control would get 30K and experimental
> > > would get 30K over the same month and same days, with same random
> > > mix of calls from demographics (same mix of area codes and MSA's
> > > etc). Control= existing IVR UI. Experimental = new UI.
> > >
> > >
> >
> > _____
> > Peter U. Leppik
> > CEO
> > Vocal Laboratories Inc.
> > pleppik@
> >
>

Mark,

The question you're asking is really secondary to this question: What
statistical analysis are you intending to run to test for statistical
significance?

My recommendation: If you are comparing two conditions and looking for a
difference in containment rate - which is a binary variable for each call (yes
it was contained or no it was not) - then a CHI square analysis is your best
bet.  The good news is that there are many CHI square calculators online (e.g.
http://people.ku.edu/~preacher/chisq/chisq.htm).  The calculator will tell you
if your results are statistically significant. As for your original question, if
you did not have enough volume into the two apps, then that will be reflected in
the CHI result.

Todd is absolutely correct. 95% is an acceptable confidence level in academic
circles.  So if the CHI value you receive comes back with a confidence level of
anything equal to or lower than 0.05 (i.e. the result was less than 5% likely
due to chance), then you have a statistically significant result.

Direct email me if you have any questions.

Good luck!

-Jon Bloom
SpeechCycle




--- In vuids@yahoogroups.com, "Mark" <webb.m@...> wrote:
>
> That's what I thought. I just wanted to get the group consensus.
>
> Do I calculate it on the total calls presented (60K) or the total calls sent
to each group (30K to control and 30K to experimental)? Looks like you've
calculated it based on the 30K per group not the total 60K.
>
> I am rounding, but it's 1/173 using 30K calls? That's 0.0058 (again rounding).
A half of a a percent.
>
> --- In vuids@yahoogroups.com, Peter Leppik <pleppik@> wrote:
> >
> > The question you're asking basically boils down to what's the margin
> > of sampling error.
> >
> > The rule of thumb for sampling error is 1/sqrt(n) where n is the
> > number of trials (in this case calls).  With 30,000 calls, your
> > sampling error will be about 0.58 percentage points--so if the
> > difference is containment is more than this, you can be fairly
> > confident that it's statistically meaningful.
> >
> >        -Peter
> >
> > On Nov 11, 2009, at 8:22 AM, Mark wrote:
> >
> > > If I am setting up a control UI vs an experimental UI to measure
> > > results, how many calls represent an acceptable volume to analyze
> > > results (results=containment) and make a decision? Would 30,000
> > > calls per server set (sets = control vs experimental) over a one
> > > month period be adequate? So control would get 30K and experimental
> > > would get 30K over the same month and same days, with same random
> > > mix of calls from demographics (same mix of area codes and MSA's
> > > etc). Control= existing IVR UI. Experimental = new UI.
> > >
> > >
> >
> > _____
> > Peter U. Leppik
> > CEO
> > Vocal Laboratories Inc.
> > pleppik@
> >
>

The measure is containment. Measuring containment in a control set of calls
(existing app, if you will) and measuring containment in an experimental app.
Sending 30,000 calls to each group over a one month period. The calc I was using
and that you confirmed is that - if measured difference is greater than .58%
(about 1/2 of 1%) then with that volume of calls the delta in containment is
statistically reliable, and reliable enough to make decisions on.

Question - I want to confirm another of my assumptions - when using the .58: is
the .58 used as a threshold to measure the ratio of the two containment rates or
a gross threshold measure of the delta. Ex: if control containment rate = 20%,
and two sample sizes = 30K each, and experimental group shows a 23% containment
rate, is the .58 used as threshold for 20/23 or 23-20?

--- In vuids@yahoogroups.com, Peter Leppik <pleppik@...> wrote:
>
> Mark:
>
> I'm assuming that you're measuring some variable in each of the two
> populations, then comparing the two measurements against each other.
>
> In this situation, the margin of error applies independently to each
> of the two populations and the two measurements could (in principle)
> have different margins of error; the correct way to calculate the
> margin of error is from the size of each sub-population.
>
> As an alternative case, if you'd chosen to have a control group of
> 45,000 calls and an experimental group of 15,000 calls, the margin of
> error in the control group would be 0.47 percentage points, and 0.82pp
> in the experimental group (there may be times when it's desirable to
> split the population other than 50/50).
>
>         -Peter
>
> On Nov 12, 2009, at 8:18 AM, Mark wrote:
>
> > That's what I thought. I just wanted to get the group consensus.
> >
> > Do I calculate it on the total calls presented (60K) or the total
> > calls sent to each group (30K to control and 30K to experimental)?
> > Looks like you've calculated it based on the 30K per group not the
> > total 60K.
> >
> > I am rounding, but it's 1/173 using 30K calls? That's 0.0058 (again
> > rounding). A half of a a percent.
> >
> > --- In vuids@yahoogroups.com, Peter Leppik <pleppik@> wrote:
> > >
> > > The question you're asking basically boils down to what's the margin
> > > of sampling error.
> > >
> > > The rule of thumb for sampling error is 1/sqrt(n) where n is the
> > > number of trials (in this case calls). With 30,000 calls, your
> > > sampling error will be about 0.58 percentage points--so if the
> > > difference is containment is more than this, you can be fairly
> > > confident that it's statistically meaningful.
> > >
> > > -Peter
> > >
> > > On Nov 11, 2009, at 8:22 AM, Mark wrote:
> > >
> > > > If I am setting up a control UI vs an experimental UI to measure
> > > > results, how many calls represent an acceptable volume to analyze
> > > > results (results=containment) and make a decision? Would 30,000
> > > > calls per server set (sets = control vs experimental) over a one
> > > > month period be adequate? So control would get 30K and
> > experimental
> > > > would get 30K over the same month and same days, with same random
> > > > mix of calls from demographics (same mix of area codes and MSA's
> > > > etc). Control= existing IVR UI. Experimental = new UI.
> > > >
> > > >
> > >
> > > _____
> > > Peter U. Leppik
> > > CEO
> > > Vocal Laboratories Inc.
> > > pleppik@
> > >
> >
> >
>
> _____
> Peter U. Leppik
> CEO
> Vocal Laboratories Inc.
> pleppik@...
>

http://careers.bankofamerica.com/JobDetails.aspx?SearchPage=Sp&JobId=0900045741&\
stateid=-1&cityid=-1&jobareas=-1|-1&keywords=&src=JB-11684

US-AZ-Chandler
Description

Responsibilities include:

• Consult with clients on requirements, design call flows, write
prompts, develop grammars, coach voice talents, and conduct usability
tests

• Take a leadership role in devising and implementing innovative
approaches to voice application design and development

• Contribute to the development of a world class voice user interface
design practice



Position requirements:

• Solid knowledge of industry best practices in voice user interface
design, solid experience crafting voice user interfaces for speech
recognition applications

• Excellent communication skills (both written and verbal), including
the ability to be both assertive and persuasive with both clients and
colleagues

• An understanding of branding in the voice environment, including
persona, audio logos, and other non-verbal messaging

• Team player with excellent initiative, follow-through, proven
ability to work effectively with internal and external customers and
manage time allocated among multiple projects

• Passion for excellence in voice user interface design





Responsible for the day-to-day resolution of complex problems and the
execution of complex transactions for multiple sites or larger
business unit(s), including research. Leads in the design, development
and implementation of complex products, systems and services in an
operations environment. Manages projects and direct activities of a
team related to special initiatives of an operations nature. Functions
as the technical expert in their assigned area. Major accountabilities
are based on individual expertise and capabilities. Provides work
direction, guidance and expertise to less experienced associates and
provides training to associates and Business Partners on new and
complex initiatives. Builds strong relationships with Business
Partners across LOBs. Has in-depth understanding of the business
unit's operations processes and implications on other groups within
the operations function and other divisions within the company.
Knowledge acquired through increasingly responsible operations
analysis work. Clearly recognized as a content expert by peers.
Bachelor degree or equivalent preferred. Individual typically has 5-7
years of experience.


Qualifications


Required Skills :



• 5 to 7 years experience in design, coding, and testing of IVR speech
and DTMF applications

• 3 to 5+ years IVR experience with technologies including web
service, SQL, CTI, VXML, and XML, Visio

• Experience working with a variety of call routing platforms, to
include, Avaya, Genesys, Nortel, Symposium, and Aspect.



Desired Skills :

• Proficiency with GSL, Nuance Voice Platform (NVP)

• A degree in cognitive science, linguistics, human factors, product
design, or related field

• Nuance certification and/or training in dialog design, grammar writing,
tunin

Hi,



I have a need to be able to take some simple text and generate a TTS mp3
from it.  I figure the easiest and cheapest is to find a VoiceXML host,
write a simple VoiceXML page that takes the text and says it using
prompt tag.  My quandry is does anyone know if I can save the resultnt
adio to MP3 using just the VoiceXML and can I find someone who can host
this?  I don't need any telephony, just an inexpensive TTS to MP3
solution.  I just don't want to license and engine, install it and write
code against it and figured I could front end it with VoiceXML.



Thoughts?



--


Dave Reich

Hi,

I have a need to be able to take some simple text and generate a TTS mp3 from
it.  I figure the easiest and cheapest is to find a VoiceXML host, write a
simple VoiceXML page that takes the text and says it using prompt tag.  My
quandry is does anyone know if I can save the resultant audio to MP3 using just
the VoiceXML and can I find someone who can host this?  I don't need any
telephony, just an inexpensive TTS solution where I can save the spoken audio to
MP3 (or au or....).  I just don't want to license an engine, install it and
write code against it and figured I could front end it with VoiceXML, post it
and capture the audio output.

Thoughts?

--

Dave Reich

[recruiter] IVR Business Analyst in San Francisco, CA

http://seeker.dice.com/jobsearch/servlet/JobSearch?op=101&dockey=xml/7/1/71b4d29\
ea81fce52b613d27384fd1906@endecaindex&c=1&source=20

M Squared Consulting, Inc. is seeking an IVR business analyst for a
San Francisco based client to be involved with a large brokerage
integration project.

Responsibilities
Provide business analyst expertise for the contact center serving
brokerage clients. The M Squared Business Analyst will work to analyze
and optimize the IVR based on a planned upgrade as well as functional
changes to support the merger of two large financial services
institutions.

Major responsibilities include:
* Analyzing and evaluating current IVR process flows specifically for
Money Movement functionality
* Leading cross functional team to understand and capture business
requirements that are used in communicating with the technology
partner
* Developing business case and making recommendations to management
team on proposed changes to the IVR to support Money Movement services
to brokerage clients including the various bridge solutions for Money
Movement service during the integration
* Analyze IVR usage to drive future functional changes and reporting
needs based on current client needs and additional requirements for
the merged company
Consultant will start upon selection. This engagement will be full
time and will be conducted on site in downtown San Francisco. The
client expects this engagement to be at least 12 to 14 months.
Deliverables
Deliverables are expected to include, but not be limited to:
* Capture and document business requirements for the IVR project for
planned new release upgrade and merger related changes
* Develop business case and recommendations for changes to IVR
* Proposition of future functional changes and reporting needs
Industry Experience
* Financial Services required
* Background in brokerage is preferred but not required
* Functional Experience
* Experience providing business analysis for IVR technology, systems
and/or projects
* 5 to 7 years of experience in business process analysis, analytics
and requirements development
* Solid understanding of technologies and processes that support the
IVR including VUI (Voice user interface) best practices, CTI (Computer
Telephony Integration) and telephony
* Understanding of technologies and processes that support the
business including business process management, workflow documentation
and analysis

Personality/Other Attributes
* Experience working in cross-functional teams and the ability to
collaborate with key stakeholders across functions
* Proven ability to influence across the organization at multiple
levels and drive results
* Ability to communicate clearly (excellent verbal and written)
Nancy DeMaagd
M Squared Consulting, Inc.
2121 Rosecrans Avenue
Suite 3350
El Segundo, CA 90245
Phone: (310) 333-8900
Fax: (310) 333-8905
Web: http://www.msquared.com

Hey Dirk; how are you? Thanks! I was not aware of this. I'll look at it
today. I just fgured VoiceXML would be the easiest path, with a couple
of lines, and capture the TTS. You can see it's a pretty simple idea I'm
after, just trying to get the cheapest overhead solution ;-)

Thanks!

Dave Reich



Dr. Dirk Schnelle-Walka wrote:
>
> Hi,
>
> I am not sure, if you REALLY need VoiceXML for this. Do you know Mary
> http://mary.dfki.de <http://mary.dfki.de> ?
>
> It is open source and you will be able to generate MP3 files from it.
>
> hth
>
> Dirk
>
> *From:* vuids@yahoogroups.com [mailto:vuids@yahoogroups.com] *On
> Behalf Of *David
> *Sent:* Tuesday, November 17, 2009 2:46 AM
> *To:* vuids@yahoogroups.com
> *Subject:* [vuids] VoiceXML TTS -> mp3?
>
> Hi,
>
> I have a need to be able to take some simple text and generate a TTS
> mp3 from it. I figure the easiest and cheapest is to find a VoiceXML
> host, write a simple VoiceXML page that takes the text and says it
> using prompt tag. My quandry is does anyone know if I can save the
> resultant audio to MP3 using just the VoiceXML and can I find someone
> who can host this? I don't need any telephony, just an inexpensive TTS
> solution where I can save the spoken audio to MP3 (or au or....). I
> just don't want to license an engine, install it and write code
> against it and figured I could front end it with VoiceXML, post it and
> capture the audio output.
>
> Thoughts?
>
> --
>
> Dave Reich
>
>

if you only need a couple of audio, Nuance has a demo section for Realspeak that
lets you save the generated TTS output as a file. 
http://www.nuance.com/realspeak/

Works out really well.

Regards,

Mark

--- In vuids@yahoogroups.com, David Reich <merlindave@...> wrote:
>
> Hey Dirk; how are you? Thanks! I was not aware of this. I'll look at it
> today. I just fgured VoiceXML would be the easiest path, with a couple
> of lines, and capture the TTS. You can see it's a pretty simple idea I'm
> after, just trying to get the cheapest overhead solution ;-)
>
> Thanks!
>
> Dave Reich

Hi,

I am not sure, if you REALLY need VoiceXML for this. Do you know Mary
http://mary.dfki.de ?
It is open source and you will be able to generate MP3 files from it.

hth
Dirk

--- In vuids@yahoogroups.com, "David" <merlindave@...> wrote:
>
> Hi,
>
> I have a need to be able to take some simple text and generate a TTS mp3 from
it.  I figure the easiest and cheapest is to find a VoiceXML host, write a
simple VoiceXML page that takes the text and says it using prompt tag.  My
quandry is does anyone know if I can save the resultant audio to MP3 using just
the VoiceXML and can I find someone who can host this?  I don't need any
telephony, just an inexpensive TTS solution where I can save the spoken audio to
MP3 (or au or....).  I just don't want to license an engine, install it and
write code against it and figured I could front end it with VoiceXML, post it
and capture the audio output.
>
> Thoughts?
>
> --
>
> Dave Reich
>

Thanks, Mark, but it would be for a low/med volume production
environment. Picture a web site where someone enters a word or sentence
(fragment) and I get a TTSed MP3 back. Hoping for a low cost hosted
alternative...

Dave Reich



markwstallings wrote:
>
> if you only need a couple of audio, Nuance has a demo section for
> Realspeak that lets you save the generated TTS output as a file.
> http://www.nuance.com/realspeak/ <http://www.nuance.com/realspeak/>
>
> Works out really well.
>
> Regards,
>
> Mark
>
> --- In vuids@yahoogroups.com <mailto:vuids%40yahoogroups.com>, David
> Reich <merlindave@...> wrote:
> >
> > Hey Dirk; how are you? Thanks! I was not aware of this. I'll look at it
> > today. I just fgured VoiceXML would be the easiest path, with a couple
> > of lines, and capture the TTS. You can see it's a pretty simple idea
> I'm
> > after, just trying to get the cheapest overhead solution ;-)
> >
> > Thanks!
> >
> > Dave Reich
>
>

Over the past year, Vocalabs conducted experiments in voice user interface
design in partnership with three different companies. We presented the results
at the 2009 SpeechTEK conference and wrote about them in our last newsletter.

This was a hugely successful undertaking to improve the state of the art in VUI
design. It's hard to get speech vendors to spend money to perform carefully
controlled experiments and publish the results for the benefit of the community
at large; it's even harder to get individual clients to take such a civic-minded
approach for their own projects.

Nevertheless, most of us can agree that the speech industry benefits greatly
from a more rigorous, open, and transparent knowledge base of VUI design
practices.  This leads to better designs, more user acceptance, and greater
project success.

VocaLabs is interested in seeing better design (and more emphasis on testing, of
course), and with the success of these experiments last year we're going to step
up to the plate again.  VocaLabs will perform usability surveys for free for any
designer or company testing VUI design practices for the benefit of the
community as a whole.

Here are the criteria:

	 • You have to have a well-designed experiment testing a design practice of
general interest, not a question relevant to only a particular design
	 • You have to commit to analyzing the data and publicizing it to the VUI design
community as a whole.
	 • You have to find someone to host the test application(s) (we can't do that
ourselves).
	 • The raw data will be made available to all interested parties in the VUI
design community (peer review and alternate analysis are especially welcome).
	 • VocaLabs has to be identified as a sponsor of the research in all
publication, presentations, etc.

If you're interested, let me know.

_____
Peter U. Leppik
CEO
Vocal Laboratories Inc.
pleppik@...

Hi,
I have a customer who would like a list of all acceptable grammar inputs for
their VUI app.  Does anyone know of a good grxml "sentence-generation" tool that
will spit out a list of items from a grxml file?
Thanks!

That will be fun.
I hope they don't plan on printing it out or they may lay waste to a
small forest somewhere.

In any non-trivial (i.e. real) grammar, there are usually enough
allowances for variant phraseology and disfluencies that there will
probably be tens of thousands of potential 'sentences', all varying by
small deviations of stop words.

Both Nuance8 and Nuance9 (which is nothing like 8) have tools to
generate sentences from grxml grammars.

Nuance8 has a tool called 'generate' (Funnily enough).

Nuance9 has 'parseTool' with the '-gen_sentences' option.

You should be able to get evaluation versions of either engine if you
are using some other engine that does not have such a feature. (Though
Nuance will probably be unhappy with such a suggestion...)


Happy generating and forest destroying!


-----Original Message-----
From: vuids@yahoogroups.com [mailto:vuids@yahoogroups.com] On Behalf Of
chillable_bend
Sent: Wednesday, 25 November 2009 11:00 AM
To: vuids@yahoogroups.com
Subject: [vuids] grxml output generator

Hi,
I have a customer who would like a list of all acceptable grammar inputs
for their VUI app.  Does anyone know of a good grxml
"sentence-generation" tool that will spit out a list of items from a
grxml file?
Thanks!



------------------------------------

Yahoo! Groups Links

I cannot agree more with Peter. Even simple grammars can lead to a large (if not
infinite) number of sentences. Generating a highly representative set of
sentences from a grammar is a non-trivial task.

In addition to the Nuance tools, you can also try NuGram
(http://nugram.nuecho.com). The Basic Edition currently supports the generation
of all sentences (which is ok for very simple grammars), while the Professional
Edition provides strategies for generating sentences in many different ways.
NuGram supports grammars for all major speech recognition engines.

--- In vuids@yahoogroups.com, "Peter Nann" <peter.nann@...> wrote:
>
> That will be fun.
> I hope they don't plan on printing it out or they may lay waste to a
> small forest somewhere.
>
> In any non-trivial (i.e. real) grammar, there are usually enough
> allowances for variant phraseology and disfluencies that there will
> probably be tens of thousands of potential 'sentences', all varying by
> small deviations of stop words.
>
> Both Nuance8 and Nuance9 (which is nothing like 8) have tools to
> generate sentences from grxml grammars.
>
> Nuance8 has a tool called 'generate' (Funnily enough).
>
> Nuance9 has 'parseTool' with the '-gen_sentences' option.
>
> You should be able to get evaluation versions of either engine if you
> are using some other engine that does not have such a feature. (Though
> Nuance will probably be unhappy with such a suggestion...)
>
>
> Happy generating and forest destroying!

Hi,
I have a customer who would like a list of all acceptable grammar inputs for their VUI app. Does anyone know of a good grxml "sentence-generation" tool that will spit out a list of items from a grxml file?
Thanks!

and... of course if there are any loops in the grammar, the list would be
infinite, and they hopefully would understand your hesitance to give them a
complete list :-)

maybe they just want a sample to give them an idea?

--- In vuids@yahoogroups.com, chillable_bend <no_reply@...> wrote:
>
> Hi,
> I have a customer who would like a list of all acceptable grammar inputs for
their VUI app.  Does anyone know of a good grxml "sentence-generation" tool that
will spit out a list of items from a grxml file?
> Thanks!
>

Thanks for the responses, everyone.  To be clear, they don't need a list of ALL
inputs, just a subset.  Their purpose is to provide the output to their UAT
group so they can test some iterations other than what is directly prompted for.
(Thankfully this is a directed dialogue app with rather simple menus, they don't
require the output of any unconstrained digit/alpha grammars, and they DON'T
want to print the output, to my knowledge)

I've used the Nuance8 tool and haven't been crazy about the results so was
looking for recommendations on something that might be a bit cleaner with more
configurable options as to what gets generated.  I was looking at the Nugram
tool that one poster mentioned so maybe I'll give that a shot.  If anyone has
direct experience with it and wants to comment on its usefulness, that would be
much appreciated.

Thanks for the input—it's good to know I'm not the only one who thinks this
customer is insane (but hey, they're paying for the effort).

Green Dot Corporation is looking for an experienced VUI/VoiceXML, fulltime
developer.
Must reside or be willing to relocate to Los Angeles area.
We offer competitive compensation and benefits.
Please contact me via email if you or someone you know might be interested.
Here is a brief job description. Complete job description available upon
request.
• Design, create and implement effective, IVR (telephone) callflows with high
usability.
• Analyze, code, test, and debug IVR programs and applications.
• Design and implement solutions using VoiceXML, XSLT, Javascript, that
integrate the IVR with supporting service and database layers.
• Tune speech applications and recommend improvements
• Troubleshoot and analyze IVR callflow applications
• Must be fluent in English and comfortable with language as a medium for
communication
• Minimum of 3 years computer software programming experience. More
specifically, VXML, service-oriented architecture, object-oriented design, .NET,
XSLT, Javascript and C# experience are all a plus, but not absolutely required.
• Excellent communication skills.
• Excellent problem-solving skills.
• Excellent interpersonal skills with the ability to develop cohesive working
relationships with internal and external clients.
• Strong technical, logical, and analytical skills.

The Nuance8 tool does allow you to avoid expanding inconsequential
sub-grammars like filler, etc.
This _might_ get you better mileage.


Here the Nuance8 option anyway:

-dont_expand grammar

     Use this option one or more times to specify subgrammars within the
grammar specified with the -grammar option that should NOT be expanded
during sentence generation.




However, in this case I think your best tool is a human.
For each choice in your grammars, pick 5 or 6 phrases that are DEAD
CERTAINTIES that callers will say.
  - Give that list to the testers.

That way you reduce the chances of the COMMON problem of testers picking
phrases that people _could_ say, but are extremely, extremely unlikely
to occur in reality.

I like to use this advice:
"Don't think of phrases people _could_ say, think of phrases that people
_would_ say. There is a difference - people _could_ say anything!. Would
you bet a month's salary on this phrase occurring from a real caller? If
not, then it's NOT a valid test phrase."




-----Original Message-----
From: vuids@yahoogroups.com [mailto:vuids@yahoogroups.com] On Behalf Of
chillable_bend
Sent: Thursday, 26 November 2009 5:22 AM
To: vuids@yahoogroups.com
Subject: [vuids] Re: grxml output generator

Thanks for the responses, everyone.  To be clear, they don't need a list
of ALL inputs, just a subset.  Their purpose is to provide the output to
their UAT group so they can test some iterations other than what is
directly prompted for.  (Thankfully this is a directed dialogue app with
rather simple menus, they don't require the output of any unconstrained
digit/alpha grammars, and they DON'T want to print the output, to my
knowledge)

I've used the Nuance8 tool and haven't been crazy about the results so
was looking for recommendations on something that might be a bit cleaner
with more configurable options as to what gets generated.  I was looking
at the Nugram tool that one poster mentioned so maybe I'll give that a
shot.  If anyone has direct experience with it and wants to comment on
its usefulness, that would be much appreciated.

Thanks for the input-it's good to know I'm not the only one who thinks
this customer is insane (but hey, they're paying for the effort).




------------------------------------

Yahoo! Groups Links