Posted: Tue Oct 31, 2006 1:24 pm Post subject: [asterisk-speech-rec] Re: Speech Recognition / 1.4beta3 +Lum
I purchased one Lumenvox for Asterisk license for the test lab -- to familiarize myself with it should a client ask if we could do a Speech Rec application.
After wrestling a week with the install, including changing OS versions and upgrading gcc libraries, I finally got it working.
Now, I'm getting way less than 50% accuracy on a simple yes/no grammar. I know it is not an Asterisk issue because when I play back the recorded "yes" or "no"
thru the Speech Tuner, it does not find the phrase in the grammar, even though it is clearly parsed as a "yes" or "no" by the SR engine.
So, after seeing how little traffic there is on this mailing list, and being quoted support rates by Lumenvox for additional assistance, I have decided to put this
product away for now. Perhaps if I have an actual production need I'll be willing to pay for support. But for now, back to DTMF-land.
From:asterisk-speech-rec-bounces@lists.digium.com [mailto:asterisk-speech-rec-bounces@lists.digium.com] On Behalf Of Gerd Graumann
Sent: Monday, October 30, 2006 9:37 PM
To: Discussing of speech recognition in Asterisk
Subject: RE: [asterisk-speech-rec] Re: Speech Recognition / 1.4beta3 +LumenVox CRM:00050623
In regards to the speech recognition related part of your question, a simple Yes/No grammar should provide you with good results.
We have recently completed a speech engine benchmark test and achieved even without "tuning" a high accuracy rate.
Please keep in mind, that the published results can be even better, when some tuning is applied to speech applications (i.e. for the name and department test)
I sent this question off to KP Fleming in the 1st instance as I was under the impression that the Speech-Rec list is more technical than commercial.
KPF has suggested that I post and seek the opinions of the community (not a bad idea at all). So here’s a copy of my question:
- - - - - - - - - - - - - - -
Kevin,
Could you advise whether it is too early for commercial implementation of 1.4beta3 and LumenVox?
We have a client who runs a marketing service sending out 500,000 faxes a week worldwide. At times the marketing list information includes up to 20% voice answers instead of fax answers.
The client pays 3c per fax and would be happy to pay this for a voice recorded announcement style message (i.e. autodialing to an existing customer list and playing a pre-recorded message).
We can do the autodialing and message announcement however, currently, at the conclusion to the message we would ask the caller to “press 1 to continue” and at that point we would connect the customer to an agent.
We’d like to:
<![if !supportLists]>- <![endif]>auto detect a fax machine or voice;
<![if !supportLists]>- <![endif]>send a fax if fax tone detected;
<![if !supportLists]>- <![endif]>if voice detected announce a simple speech recognition option which includes the following“…if you would like to speak to one of our friendly sales staff or hear more about this great offer, just say ‘yes’ now…..”;
Before we allocate our engineers to this task I wanted to check to see if this is not only possible, but whether it would be considered stable enough to be used in production.
Posted: Tue Oct 31, 2006 2:09 pm Post subject: [asterisk-speech-rec] Re: Speech Recognition / 1.4beta3 +Lum
Hi Adam-
I participated in the early Lumenvox-Asterisk beta program and, after
the little hassles of upgrading Fedora Core, I was able to get
everything working. I experienced very good results in simply decoding
the numbers 0-9 and a few command words - the accuracy was impressive. I
did have problems with accuracy at first, mostly false decoding, but I
was able to resolve this by cranking the input audio level down a bit as
I recall. Maybe in your case playing with the levels will help?
I purchased one Lumenvox for Asterisk license for the test lab -- to
familiarize myself with it should a client ask if we could do a Speech
Rec application.
After wrestling a week with the install, including changing OS
versions and upgrading gcc libraries, I finally got it working.
Now, I'm getting way less than 50% accuracy on a simple yes/no
grammar. I know it is not an Asterisk issue because when I play back
the recorded "yes" or "no"
thru the Speech Tuner, it does not find the phrase in the grammar,
even though it is clearly parsed as a "yes" or "no" by the SR engine.
So, after seeing how little traffic there is on this mailing list, and
being quoted support rates by Lumenvox for additional assistance, I
have decided to put this
product away for now. Perhaps if I have an actual production need I'll
be willing to pay for support. But for now, back to DTMF-land.
------------------------------------------------------------------------
*From:* asterisk-speech-rec-bounces@lists.digium.com
[mailto:asterisk-speech-rec-bounces@lists.digium.com] *On Behalf Of
*Gerd Graumann
*Sent:* Monday, October 30, 2006 9:37 PM
*To:* Discussing of speech recognition in Asterisk
*Subject:* RE: [asterisk-speech-rec] Re: Speech Recognition / 1.4beta3
+LumenVox CRM:00050623
In regards to the speech recognition related part of your question, a
simple Yes/No grammar should provide you with good results.
We have recently completed a speech engine benchmark test and achieved
even without "tuning" a high accuracy rate.
http://www.lumenvox.com/products/speech_engine/accuracyStudy/
Please keep in mind, that the published results can be even better,
when some tuning is applied to speech applications (i.e. for the name
and department test)
Best Regards,
Gerd Graumann
Director of Business Development
P: 877-977-0707, just say "Gerd"
F: 858-707-7072
Gerd@LumenVox.com <mailto:Gerd@LumenVox.com>
www.lumenvox.com <http://www.lumenvox.com/>
<http://www.lumenvox.com/>
Tell us what is important to you! Take our product development survey:
http://www.LumenVox.com/survey/srDevelopment/index.asp
<http://www.lumenvox.com/survey/srDevelopment/index.asp>
------------------------------------------------------------------------
*From:* asterisk-speech-rec-bounces@lists.digium.com
[mailto:asterisk-speech-rec-bounces@lists.digium.com] *On Behalf Of
*Craig Lawrence
*Sent:* Sunday, October 29, 2006 1:25 AM
*To:* asterisk-speech-rec@lists.digium.com
*Subject:* [asterisk-speech-rec] Re: Speech Recognition / 1.4beta3 +
LumenVox
Hi All
I sent this question off to KP Fleming in the 1^st instance as I was
under the impression that the Speech-Rec list is more technical than
commercial.
KPF has suggested that I post and seek the opinions of the community
(not a bad idea at all). So here’s a copy of my question:
- - - - - - - - - - - - - - -
Kevin,
Could you advise whether it is too early for commercial implementation
of 1.4beta3 and LumenVox?
We have a client who runs a marketing service sending out 500,000
faxes a week worldwide. At times the marketing list information
includes up to 20% voice answers instead of fax answers.
The client pays 3c per fax and would be happy to pay this for a voice
recorded announcement style message (i.e. autodialing to an existing
customer list and playing a pre-recorded message).
We can do the autodialing and message announcement however, currently,
at the conclusion to the message we would ask the caller to “press 1
to continue” and at that point we would connect the customer to an agent.
We’d like to:
- auto detect a fax machine or voice;
- send a fax if fax tone detected;
- if voice detected announce a simple speech recognition option which
includes the following“…if you would like to speak to one of our
friendly sales staff or hear more about this great offer, just say
‘yes’ now…..”;
Before we allocate our engineers to this task I wanted to check to see
if this is not only possible, but whether it would be considered
stable enough to be used in production.
Posted: Tue Oct 31, 2006 2:31 pm Post subject: [asterisk-speech-rec] Re: Speech Recognition / 1.4beta3 +Lum
Thanks, I'll give it a try. What's strange is that the SRE properly
sees the recording as "yes" or "no" but then does not find it in the
grammar. It's not like it simply doesn't understand what I'm saying.
-----Original Message-----
From: asterisk-speech-rec-bounces@lists.digium.com
[mailto:asterisk-speech-rec-bounces@lists.digium.com] On Behalf Of Scott
Stingel
Sent: Tuesday, October 31, 2006 9:05 AM
To: Discussing of speech recognition in Asterisk
Subject: Re: [asterisk-speech-rec] Re: Speech Recognition / 1.4beta3
+LumenVoxCRM:00050623
Hi Adam-
I participated in the early Lumenvox-Asterisk beta program and, after
the little hassles of upgrading Fedora Core, I was able to get
everything working. I experienced very good results in simply decoding
the numbers 0-9 and a few command words - the accuracy was impressive. I
did have problems with accuracy at first, mostly false decoding, but I
was able to resolve this by cranking the input audio level down a bit as
I recall. Maybe in your case playing with the levels will help?
I purchased one Lumenvox for Asterisk license for the test lab -- to
familiarize myself with it should a client ask if we could do a Speech
Quote:
Rec application.
After wrestling a week with the install, including changing OS
versions and upgrading gcc libraries, I finally got it working.
Now, I'm getting way less than 50% accuracy on a simple yes/no
grammar. I know it is not an Asterisk issue because when I play back
the recorded "yes" or "no"
thru the Speech Tuner, it does not find the phrase in the grammar,
even though it is clearly parsed as a "yes" or "no" by the SR engine.
So, after seeing how little traffic there is on this mailing list, and
Quote:
being quoted support rates by Lumenvox for additional assistance, I
have decided to put this product away for now. Perhaps if I have an
actual production need I'll be willing to pay for support. But for
now, back to DTMF-land.
----------------------------------------------------------------------
--
*From:* asterisk-speech-rec-bounces@lists.digium.com
[mailto:asterisk-speech-rec-bounces@lists.digium.com] *On Behalf Of
*Gerd Graumann
*Sent:* Monday, October 30, 2006 9:37 PM
*To:* Discussing of speech recognition in Asterisk
*Subject:* RE: [asterisk-speech-rec] Re: Speech Recognition / 1.4beta3
+LumenVox CRM:00050623
In regards to the speech recognition related part of your question, a
simple Yes/No grammar should provide you with good results.
We have recently completed a speech engine benchmark test and achieved
Quote:
even without "tuning" a high accuracy rate.
http://www.lumenvox.com/products/speech_engine/accuracyStudy/
Please keep in mind, that the published results can be even better,
when some tuning is applied to speech applications (i.e. for the name
and department test) Best Regards, Gerd Graumann Director of Business
Development
P: 877-977-0707, just say "Gerd"
F: 858-707-7072
Gerd@LumenVox.com <mailto:Gerd@LumenVox.com> www.lumenvox.com
<http://www.lumenvox.com/> <http://www.lumenvox.com/>
Tell us what is important to you! Take our product development survey:
http://www.LumenVox.com/survey/srDevelopment/index.asp
<http://www.lumenvox.com/survey/srDevelopment/index.asp>
----------------------------------------------------------------------
--
*From:* asterisk-speech-rec-bounces@lists.digium.com
[mailto:asterisk-speech-rec-bounces@lists.digium.com] *On Behalf Of
*Craig Lawrence
*Sent:* Sunday, October 29, 2006 1:25 AM
*To:* asterisk-speech-rec@lists.digium.com
*Subject:* [asterisk-speech-rec] Re: Speech Recognition / 1.4beta3 +
LumenVox
Hi All
I sent this question off to KP Fleming in the 1^st instance as I was
under the impression that the Speech-Rec list is more technical than
commercial.
KPF has suggested that I post and seek the opinions of the community
(not a bad idea at all). So here's a copy of my question:
- - - - - - - - - - - - - - -
Kevin,
Could you advise whether it is too early for commercial implementation
Quote:
of 1.4beta3 and LumenVox?
We have a client who runs a marketing service sending out 500,000
faxes a week worldwide. At times the marketing list information
includes up to 20% voice answers instead of fax answers.
The client pays 3c per fax and would be happy to pay this for a voice
recorded announcement style message (i.e. autodialing to an existing
customer list and playing a pre-recorded message).
We can do the autodialing and message announcement however, currently,
Quote:
at the conclusion to the message we would ask the caller to "press 1
to continue" and at that point we would connect the customer to an
agent.
Quote:
We'd like to:
- auto detect a fax machine or voice;
- send a fax if fax tone detected;
- if voice detected announce a simple speech recognition option which
includes the following"...if you would like to speak to one of our
friendly sales staff or hear more about this great offer, just say
'yes' now.....";
Before we allocate our engineers to this task I wanted to check to see
Quote:
if this is not only possible, but whether it would be considered
stable enough to be used in production.
-----Original Message-----
From: asterisk-speech-rec-bounces@lists.digium.com [mailto:asterisk-
speech-rec-bounces@lists.digium.com] On Behalf Of Scott Stingel
Sent: Tuesday, 31 October 2006 9:05 AM
To: Discussing of speech recognition in Asterisk
Subject: Re: [asterisk-speech-rec] Re: Speech Recognition / 1.4beta3
+LumenVoxCRM:00050623
Hi Adam-
I participated in the early Lumenvox-Asterisk beta program and, after
the little hassles of upgrading Fedora Core, I was able to get
everything working. I experienced very good results in simply decoding
the numbers 0-9 and a few command words - the accuracy was impressive.
I
Quote:
did have problems with accuracy at first, mostly false decoding, but I
was able to resolve this by cranking the input audio level down a bit
as
Quote:
I recall. Maybe in your case playing with the levels will help?
Adam Robins wrote:
> I purchased one Lumenvox for Asterisk license for the test lab -- to
> familiarize myself with it should a client ask if we could do a
Speech
Quote:
> Rec application.
> After wrestling a week with the install, including changing OS
> versions and upgrading gcc libraries, I finally got it working.
> Now, I'm getting way less than 50% accuracy on a simple yes/no
> grammar. I know it is not an Asterisk issue because when I play back
> the recorded "yes" or "no"
> thru the Speech Tuner, it does not find the phrase in the grammar,
> even though it is clearly parsed as a "yes" or "no" by the SR
engine.
Quote:
> So, after seeing how little traffic there is on this mailing list,
and
Quote:
> being quoted support rates by Lumenvox for additional assistance, I
> have decided to put this
> product away for now. Perhaps if I have an actual production need
I'll
Quote:
> be willing to pay for support. But for now, back to DTMF-land.
>
>
------------------------------------------------------------------------
Quote:
> *From:* asterisk-speech-rec-bounces@lists.digium.com
> [mailto:asterisk-speech-rec-bounces@lists.digium.com] *On Behalf Of
> *Gerd Graumann
> *Sent:* Monday, October 30, 2006 9:37 PM
> *To:* Discussing of speech recognition in Asterisk
> *Subject:* RE: [asterisk-speech-rec] Re: Speech Recognition /
1.4beta3
Quote:
> +LumenVox CRM:00050623
>
> In regards to the speech recognition related part of your question,
a
Quote:
> simple Yes/No grammar should provide you with good results.
> We have recently completed a speech engine benchmark test and
achieved
Quote:
> even without "tuning" a high accuracy rate.
> http://www.lumenvox.com/products/speech_engine/accuracyStudy/
> Please keep in mind, that the published results can be even better,
> when some tuning is applied to speech applications (i.e. for the
name
Quote:
> and department test)
> Best Regards,
> Gerd Graumann
> Director of Business Development
> P: 877-977-0707, just say "Gerd"
> F: 858-707-7072
> Gerd@LumenVox.com <mailto:Gerd@LumenVox.com>
> www.lumenvox.com <http://www.lumenvox.com/>
> <http://www.lumenvox.com/>
>
> Tell us what is important to you! Take our product development
survey:
> *From:* asterisk-speech-rec-bounces@lists.digium.com
> [mailto:asterisk-speech-rec-bounces@lists.digium.com] *On Behalf Of
> *Craig Lawrence
> *Sent:* Sunday, October 29, 2006 1:25 AM
> *To:* asterisk-speech-rec@lists.digium.com
> *Subject:* [asterisk-speech-rec] Re: Speech Recognition / 1.4beta3 +
> LumenVox
>
> Hi All
>
> I sent this question off to KP Fleming in the 1^st instance as I was
> under the impression that the Speech-Rec list is more technical than
> commercial.
>
> KPF has suggested that I post and seek the opinions of the community
> (not a bad idea at all). So here's a copy of my question:
>
> - - - - - - - - - - - - - - -
>
> Kevin,
>
> Could you advise whether it is too early for commercial
implementation
Quote:
> of 1.4beta3 and LumenVox?
>
> We have a client who runs a marketing service sending out 500,000
> faxes a week worldwide. At times the marketing list information
> includes up to 20% voice answers instead of fax answers.
>
> The client pays 3c per fax and would be happy to pay this for a
voice
Quote:
> recorded announcement style message (i.e. autodialing to an existing
> customer list and playing a pre-recorded message).
>
> We can do the autodialing and message announcement however,
currently,
Quote:
> at the conclusion to the message we would ask the caller to "press 1
> to continue" and at that point we would connect the customer to an
agent.
>
> We'd like to:
>
> - auto detect a fax machine or voice;
>
> - send a fax if fax tone detected;
>
> - if voice detected announce a simple speech recognition option
which
Quote:
> includes the following"...if you would like to speak to one of our
> friendly sales staff or hear more about this great offer, just say
> 'yes' now.....";
>
> Before we allocate our engineers to this task I wanted to check to
see
Quote:
> if this is not only possible, but whether it would be considered
> stable enough to be used in production.
>
> Your comments would be appreciated.
>
> - - - - - - - - - - -
>
> Any comments / opinions would be appreciated.
>
> Regards
>
> **Craig Lawrence**
>
> www.mytel.net.au <http://www.mytel.net.au>
>
>
> --
> No virus found in this outgoing message.
> Checked by AVG Free Edition.
> Version: 7.1.408 / Virus Database: 268.13.17/505 - Release Date:
> 27/10/2006
>
>
------------------------------------------------------------------------
Quote:
>
> _______________________________________________
> --Bandwidth and Colocation provided by Easynews.com --
>
> asterisk-speech-rec mailing list
> To UNSUBSCRIBE or update options visit:
> http://lists.digium.com/mailman/listinfo/asterisk-speech-rec
>
_______________________________________________
--Bandwidth and Colocation provided by Easynews.com --
Posted: Tue Oct 31, 2006 2:44 pm Post subject: [asterisk-speech-rec] Re: Speech Recognition / 1.4beta3 +Lum
I had the exact same issue when using zaptel cards, If you have RX gain set
to just about anything over zero there is a problem with accuracy but if you
use strictly voip it seems to work really well.
-----Original Message-----
From: asterisk-speech-rec-bounces@lists.digium.com
[mailto:asterisk-speech-rec-bounces@lists.digium.com] On Behalf Of Scott
Stingel
Sent: Tuesday, October 31, 2006 9:05 AM
To: Discussing of speech recognition in Asterisk
Subject: Re: [asterisk-speech-rec] Re: Speech Recognition / 1.4beta3
+LumenVoxCRM:00050623
Hi Adam-
I participated in the early Lumenvox-Asterisk beta program and, after
the little hassles of upgrading Fedora Core, I was able to get
everything working. I experienced very good results in simply decoding
the numbers 0-9 and a few command words - the accuracy was impressive. I
did have problems with accuracy at first, mostly false decoding, but I
was able to resolve this by cranking the input audio level down a bit as
I recall. Maybe in your case playing with the levels will help?
I purchased one Lumenvox for Asterisk license for the test lab -- to
familiarize myself with it should a client ask if we could do a Speech
Rec application.
After wrestling a week with the install, including changing OS
versions and upgrading gcc libraries, I finally got it working.
Now, I'm getting way less than 50% accuracy on a simple yes/no
grammar. I know it is not an Asterisk issue because when I play back
the recorded "yes" or "no"
thru the Speech Tuner, it does not find the phrase in the grammar,
even though it is clearly parsed as a "yes" or "no" by the SR engine.
So, after seeing how little traffic there is on this mailing list, and
being quoted support rates by Lumenvox for additional assistance, I
have decided to put this
product away for now. Perhaps if I have an actual production need I'll
be willing to pay for support. But for now, back to DTMF-land.
------------------------------------------------------------------------
*From:* asterisk-speech-rec-bounces@lists.digium.com
[mailto:asterisk-speech-rec-bounces@lists.digium.com] *On Behalf Of
*Gerd Graumann
*Sent:* Monday, October 30, 2006 9:37 PM
*To:* Discussing of speech recognition in Asterisk
*Subject:* RE: [asterisk-speech-rec] Re: Speech Recognition / 1.4beta3
+LumenVox CRM:00050623
In regards to the speech recognition related part of your question, a
simple Yes/No grammar should provide you with good results.
We have recently completed a speech engine benchmark test and achieved
even without "tuning" a high accuracy rate.
http://www.lumenvox.com/products/speech_engine/accuracyStudy/
Please keep in mind, that the published results can be even better,
when some tuning is applied to speech applications (i.e. for the name
and department test)
Best Regards,
Gerd Graumann
Director of Business Development
P: 877-977-0707, just say "Gerd"
F: 858-707-7072
Gerd@LumenVox.com <mailto:Gerd@LumenVox.com>
www.lumenvox.com <http://www.lumenvox.com/>
<http://www.lumenvox.com/>
Tell us what is important to you! Take our product development survey:
http://www.LumenVox.com/survey/srDevelopment/index.asp
<http://www.lumenvox.com/survey/srDevelopment/index.asp>
------------------------------------------------------------------------
*From:* asterisk-speech-rec-bounces@lists.digium.com
[mailto:asterisk-speech-rec-bounces@lists.digium.com] *On Behalf Of
*Craig Lawrence
*Sent:* Sunday, October 29, 2006 1:25 AM
*To:* asterisk-speech-rec@lists.digium.com
*Subject:* [asterisk-speech-rec] Re: Speech Recognition / 1.4beta3 +
LumenVox
Hi All
I sent this question off to KP Fleming in the 1^st instance as I was
under the impression that the Speech-Rec list is more technical than
commercial.
KPF has suggested that I post and seek the opinions of the community
(not a bad idea at all). So here's a copy of my question:
- - - - - - - - - - - - - - -
Kevin,
Could you advise whether it is too early for commercial implementation
of 1.4beta3 and LumenVox?
We have a client who runs a marketing service sending out 500,000
faxes a week worldwide. At times the marketing list information
includes up to 20% voice answers instead of fax answers.
The client pays 3c per fax and would be happy to pay this for a voice
recorded announcement style message (i.e. autodialing to an existing
customer list and playing a pre-recorded message).
We can do the autodialing and message announcement however, currently,
at the conclusion to the message we would ask the caller to "press 1
to continue" and at that point we would connect the customer to an agent.
We'd like to:
- auto detect a fax machine or voice;
- send a fax if fax tone detected;
- if voice detected announce a simple speech recognition option which
includes the following".if you would like to speak to one of our
friendly sales staff or hear more about this great offer, just say
'yes' now...";
Before we allocate our engineers to this task I wanted to check to see
if this is not only possible, but whether it would be considered
stable enough to be used in production.
You can post new topics in this forum You can reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum