[on-asterisk] Queue Timeouts
Hi everyone, I have a call centre that processes 300-500 calls a day with multiple queues using a PRI. I am using AEL, and this is how the way I setup the queues: Queue(QUEUE-LEVEL1,rt,,,20); Queue(QUEUE-LEVEL2,t,,,1800); I am tracking down an SLA and I notice that I have calls staying in LEVEL1 on average of 26 secs... and the maximum is 32. As you can see the timeout is 20secs... but I have calls stay in LEVEL1 as long as 32 secs...has anyone run into this issue? Any advise / how do you track down your QUEUE SLA? Input/direction is appreciated. Thanks in advance. -- Francis Ballares E-mail: francis (at) voipware.ca Local: 1(647)722-5627 x.401 Toll: 1(888)312-3747 Web: www.voipware.ca - Your Canadian On-Line Source for VoIP, Network and Telephony Hardware. Note: This e-mail may contain confidential and/or privileged information. If you are not the intended recipient (or have received this e-mail in error)please notify the sender immediately and destroy this e-mail. Any unauthorized copying, disclosure or distribution of the material in this e-mail is strictly forbidden. - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: [on-asterisk] Queue Timeouts
On 10/17/07, Francis Ballares [EMAIL PROTECTED] wrote: Hi everyone, I have a call centre that processes 300-500 calls a day with multiple queues using a PRI. I am using AEL, and this is how the way I setup the queues: Queue(QUEUE-LEVEL1,rt,,,20); Queue(QUEUE-LEVEL2,t,,,1800); I am tracking down an SLA and I notice that I have calls staying in LEVEL1 on average of 26 secs... and the maximum is 32. As you can see the timeout is 20secs... but I have calls stay in LEVEL1 as long as 32 secs...has anyone run into this issue? Any advise / how do you track down your QUEUE SLA? Input/direction is appreciated. There's a blurb about this on the wiki: http://www.voip-info.org/wiki/view/Asterisk+cmd+Queue ...Once a call is inside a queue, it is only checked to see if it is timed out every so often (depending on the value of 'timeout=' for that queue in queues.conf). This means that if queues.conf says 'timeout=10' and you call the queue with 'Queue(queuename5)', it will be ten seconds before the call times out -- j.
[on-asterisk] IT360 Asterisk and Open Telephony Conference
Hello all, Together with the help of TAUG, I'm planning to make sure that the second annual Asterisk conference in Toronto (April 7-9 at the Toronto Convention Centre) is even better than the inaugural one. I hope to be working more with Simon, Jim and anyone interested in making the conference program engaging, relevant and provocative. At this stage I'm reviewing the sessions we had last year and could certainly use some input regarding what worked, what didn't, what topics folks you'd like to hear about and which speakers you'd like to hear. Everything is open for discussion -- the intent is making a conference with subjects and speakers that people in the community, as well as those new to Asterisk and open source, want to hear. As a point of reference, here's list of the sessions that were done at the first conference. Installing Asterisk Integrating and Securing Asterisk A gentle introduction to VOIP for business Migrating from POTS to VOIP The Consumer side of VOIP Asterisk and small business Open Source VOIP War Stories What are Enterprises Doing with Asterisk? Exploring the Unknown -- what you didn't know your phone system could do Open Source Call Centres Understanding SIP Open Source VOIP for Resellers and Integrators The Future of Asterisk Open Source VOIP War Stories Any and all suggestions are appreciated. Thanks! - Evan Leibovitch - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
[on-asterisk] Speech Recogniton Reliability
Has any one implemented SPEECH RECOGNITION in Asterisk in a production environment? If not... in a test environment, what are your findings, when the same English sentence is spoken by a dozen different people with a dozen different accents/dialects in English. How accurate is the speech engine with wrong grammar spoken. How many simultaneous recognitions it can do on a P4, GHz dual core on 2 Gigs of Ram. etc. Any bench mark is appreciated. Best, Reza. - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: [on-asterisk] Speech Recogniton Reliability
I assume you just got the targeted email (aka SPAM) from Digium as well. - Original Message - From: Reza - Asterisk Enthusiast [EMAIL PROTECTED] To: TAUG asterisk@uc.org Sent: Wednesday, October 17, 2007 3:52 PM Subject: [on-asterisk] Speech Recogniton Reliability Has any one implemented SPEECH RECOGNITION in Asterisk in a production environment? If not... in a test environment, what are your findings, when the same English sentence is spoken by a dozen different people with a dozen different accents/dialects in English. How accurate is the speech engine with wrong grammar spoken. How many simultaneous recognitions it can do on a P4, GHz dual core on 2 Gigs of Ram. etc. Any bench mark is appreciated. Best, Reza. - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: [on-asterisk] Speech Recogniton Reliability
LOL! Yes and No :).I was first introduced to this through Simon's presentation a while ago... and Digium a while ago had a commercial module or something available as well for a small fee - for which I received targeted SPAM. LOL! ... wanted to hear from others about it, and we've been contracted to produce system for a survey company. The responses are basically: - Yes - No - and numbers 1 to 5. The app works in terms of punching in 1 for yes, 2 for no. and then the numbers for a scale of 1 - 5. I want to take the application for both personal and business reasons -- to the next frontier :). Speech recognition. Before I immerse myself playing with this -- I want to hear what others have done with speech. Have you tinkered with this? Cheers! Reza. - Original Message - From: Bill Sandiford [EMAIL PROTECTED] To: Reza - Asterisk Enthusiast [EMAIL PROTECTED]; TAUG asterisk@uc.org Sent: Wednesday, October 17, 2007 4:19 PM Subject: Re: [on-asterisk] Speech Recogniton Reliability I assume you just got the targeted email (aka SPAM) from Digium as well. - Original Message - From: Reza - Asterisk Enthusiast [EMAIL PROTECTED] To: TAUG asterisk@uc.org Sent: Wednesday, October 17, 2007 3:52 PM Subject: [on-asterisk] Speech Recogniton Reliability Has any one implemented SPEECH RECOGNITION in Asterisk in a production environment? If not... in a test environment, what are your findings, when the same English sentence is spoken by a dozen different people with a dozen different accents/dialects in English. How accurate is the speech engine with wrong grammar spoken. How many simultaneous recognitions it can do on a P4, GHz dual core on 2 Gigs of Ram. etc. Any bench mark is appreciated. Best, Reza. - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: [on-asterisk] Speech Recogniton Reliability
Hi Reza, I did a bit of playing with Sphynx and it was fast and accurate with small word lists. I didn't try anything as complicated as understanding the semantics of sentences. A simple word list like Yes, No, Exit, Help, one, two, three, four, five should be pretty easy. When you open up the dictionary to the full english language and full range of accents, my couch sounds like a Ohian's (someone from Ohio?) coach and my pipe sounds like a Philipino five. I think it starts to get dicey with anything but a strictly limited word list. Having said all that, I haven't put anything into production and would defer to anyone with expertise in the subject. Dave On 10/17/07, Reza - Asterisk Enthusiast [EMAIL PROTECTED] wrote: LOL! Yes and No :).I was first introduced to this through Simon's presentation a while ago... and Digium a while ago had a commercial module or something available as well for a small fee - for which I received targeted SPAM. LOL! ... wanted to hear from others about it, and we've been contracted to produce system for a survey company. The responses are basically: - Yes - No - and numbers 1 to 5. The app works in terms of punching in 1 for yes, 2 for no. and then the numbers for a scale of 1 - 5. I want to take the application for both personal and business reasons -- to the next frontier :). Speech recognition. Before I immerse myself playing with this -- I want to hear what others have done with speech. Have you tinkered with this? Cheers! Reza. - Original Message - From: Bill Sandiford [EMAIL PROTECTED] To: Reza - Asterisk Enthusiast [EMAIL PROTECTED]; TAUG asterisk@uc.org Sent: Wednesday, October 17, 2007 4:19 PM Subject: Re: [on-asterisk] Speech Recogniton Reliability I assume you just got the targeted email (aka SPAM) from Digium as well. - Original Message - From: Reza - Asterisk Enthusiast [EMAIL PROTECTED] To: TAUG asterisk@uc.org Sent: Wednesday, October 17, 2007 3:52 PM Subject: [on-asterisk] Speech Recogniton Reliability Has any one implemented SPEECH RECOGNITION in Asterisk in a production environment? If not... in a test environment, what are your findings, when the same English sentence is spoken by a dozen different people with a dozen different accents/dialects in English. How accurate is the speech engine with wrong grammar spoken. How many simultaneous recognitions it can do on a P4, GHz dual core on 2 Gigs of Ram. etc. Any bench mark is appreciated. Best, Reza. - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: [on-asterisk] Speech Recogniton Reliability
Reza, Now we've not used speech to text with asterisk, but use it as an independant custom app we have developed. Most examples of speech to text I've seen using asterisk has used Sphinx V2. It doesn't require much horse power but s limited in it's ability. I believe this is the engine that Simon used for his Adventure project. We utilize Sphinx V4 which is written in java and uses a more sophisticated recognition method HMM (Hidden Markov Models) for enhanced recognition. More info here: - http://cmusphinx.sourceforge.net/html/cmusphinx.php - http://www.speech.cs.cmu.edu/sphinx/twiki/bin/view/Sphinx4/WebHome . There is a better engine out there, the CSLU Toolkit ( http://www.cslu.ogi.edu ). It is free for non-commercial use, but for commercial use requires a US$10,000 liscencing fee. Basically the main funder is DARPA ( Defense Advanced Research Projects Agency ). They funded the Sphinx project for 5 years at CMU (Carnagie Melon). When the project ran it's life, DARPA moved the funding to OGI ( part of Oregon Health Science University ), who continues the further advancement. Now on the performance side, I'm going to guestimate based on our experience. The dictionary size and audio size are the two main driving parameters. We transcribe an average 6 minute conversation, using a limited 200 word dictionary at 7x realtime speed, now we have a larger error rate, due to the larger dictionary and that the verbage is unpredictable and has two people speaking. This is not done in realtime and we load the whole 6 minutes into ram uncompressed. This is done on a 1.67Ghz Celeron processor w 2GB ram. When processing it consumes almost 1GB. Now in your instance, you'd be feeding it very small snippets and using a very limited dictionary, say (0-9 yes no help operator) so 14 words. You should be able to run this on your targeted system, depending on how you confiure it, and the ratio of listening time VS speech2Text processing will dictate your concurrent call capability. So my guess, 20-40 (possibly more) inbound channels depending on the ratio. For the limited dictionary size, that your targeting, you should be able to achieve a less then 2% error rate with Sphinx4. A few other techniques that can improve the recognition: - one - is to have a seperate dictionary and use it to attempt to identify the ethnicity/accent of the caller, then utilize a specialized language model and dictionary for them to improve accuracy, but the negative will be higher resource costs and complexity. - two - is to build out your dictionary with alternative pronounciations of all the words in the dictionary. - eg. THREE - as TH-R-II , or as T-R-II or Tu-R-II so having thre records pointing to Three. The codes look wierd but is based on phonemes. Hope this helps a bit. Mike Reza - Asterisk Enthusiast wrote: LOL! Yes and No :).I was first introduced to this through Simon's presentation a while ago... and Digium a while ago had a commercial module or something available as well for a small fee - for which I received targeted SPAM. LOL! ... wanted to hear from others about it, and we've been contracted to produce system for a survey company. The responses are basically: - Yes - No - and numbers 1 to 5. The app works in terms of punching in 1 for yes, 2 for no. and then the numbers for a scale of 1 - 5. I want to take the application for both personal and business reasons -- to the next frontier :). Speech recognition. Before I immerse myself playing with this -- I want to hear what others have done with speech. Have you tinkered with this? Cheers! Reza. - Original Message - From: Bill Sandiford [EMAIL PROTECTED] To: Reza - Asterisk Enthusiast [EMAIL PROTECTED]; TAUG asterisk@uc.org Sent: Wednesday, October 17, 2007 4:19 PM Subject: Re: [on-asterisk] Speech Recogniton Reliability I assume you just got the targeted email (aka SPAM) from Digium as well. - Original Message - From: Reza - Asterisk Enthusiast [EMAIL PROTECTED] To: TAUG asterisk@uc.org Sent: Wednesday, October 17, 2007 3:52 PM Subject: [on-asterisk] Speech Recogniton Reliability Has any one implemented SPEECH RECOGNITION in Asterisk in a production environment? If not... in a test environment, what are your findings, when the same English sentence is spoken by a dozen different people with a dozen different accents/dialects in English. How accurate is the speech engine with wrong grammar spoken. How many simultaneous recognitions it can do on a P4, GHz dual core on 2 Gigs of Ram. etc. Any bench mark is appreciated. Best, Reza. - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL
Re: [on-asterisk] Speech Recogniton Reliability
Reza - Asterisk Enthusiast wrote: Has any one implemented SPEECH RECOGNITION in Asterisk in a production environment? If not... in a test environment, what are your findings, when the same English sentence is spoken by a dozen different people with a dozen different accents/dialects in English. How accurate is the speech engine with wrong grammar spoken. How many simultaneous recognitions it can do on a P4, GHz dual core on 2 Gigs of Ram. etc. Someone put out a mud game or something ages ago that they could play over the phone using text to speech and speech to text, I kept meaning to get round to trying it but never found the time and had forgotten about it till I saw your email -- Best regards, Duane http://www.freeauth.org - Enterprise Two Factor Authentication http://www.nodedb.com - Think globally, network locally http://www.sydneywireless.com - Telecommunications Freedom http://e164.org - Because e164.arpa is a tax on VoIP In the long run the pessimist may be proved right, but the optimist has a better time on the trip. - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]