V20 U2 Alpha - AI, Microsoft Teams & Admin Control

[ivank]

Does anyone know how "OpenAPI - Whisper" compares to Google's transcription both in terms of price and transcription quality?

From my tests I can tell that Whisper is considerably better quality-wise (at least, for English). Actual prices can be easily googled I believe.

 

[bitn2]

why the "Provider Import" (XML) on the trunk page is gone on the new alpha??
have to change "RemotePartyIDCallingPartyUserPart" and this is only possible within xml?!

Thats really bad! I can reproduce this problem as well...

 

[UCMUserAZ]

you can input below command to upgrade Update 2(do not run it for production 3CX)

echo "deb [arch=amd64 by-hash=yes signed-by=/usr/share/keyrings/3cx-archive-keyring.gpg] http://repo.3cx.com/3cx bookworm-testing main" | tee /etc/apt/sources.list.d/3cxpbx.list
apt update
apt -t bookworm-testing upgrade 3cxpbx

if upgrade failed, also can run below command:

apt remove 3cxpbx
apt install 3cxpbx=20.0.2.688

Thanks. 3cx responded and we got it fixed. Subset of systems did not have ability to see Alpha and Beta releases. Fixed now.

 

[AramS_3CX]

@wiesideluxe and @bitn2
Hi guys,
The provider import is moved to the Advanced->Templates->Provider Templates page.
You can use it from there.

 

[bitn2]

@wiesideluxe and @bitn2
Hi guys,
The provider import is moved to the Advanced->Templates->Provider Templates page.
You can use it from there.
View attachment 41959

Uff... thx.. i was a little shocked for the moment.

 

[CentrexJ]

From my tests I can tell that Whisper is considerably better quality-wise (at least, for English). Actual prices can be easily googled I believe.

Good point. I think i found it Likely USD pricing. So a lot cheaper for better. very nice upgrade.

OpenAI Whisper is $0.006/minute
Google speech to text v2 api $0.016/minute (3CX v20)
Google speech to text v1 api $0.024/minute (3CX v18)

 

[João Vitor]

The OpenAI addition for transcription and sentiment analysis is simply great! But I'm still in doubt regarding the sentiment analisys scoring, is it 1 to 5, being 1 a person visibly "angry" and 5 someone "happy"? Does it give a scoring based on both persons manner of speaking or it only considers let's say what the customer said (that's assuming it would also identify when a extension received or made a call)?

 

[Alphabetic]

Hello,

it looks like the OpenAI requirements are updated very dynamically, we will update our documentation to specify everything in detail so please bear with us. To start using OpenAI you need to set the Limits and Model Usage in your project settings. I have attached screenshot of a working configuration for those limits. Try this and let us know if this works for you

View attachment 41913

Must I get to tier 1 for this to work? Currently in testing phase. Will it work with free tier?

 

[nikolascx]

The OpenAI addition for transcription and sentiment analysis is simply great! But I'm still in doubt regarding the sentiment analisys scoring, is it 1 to 5, being 1 a person visibly "angry" and 5 someone "happy"? Does it give a scoring based on both persons manner of speaking or it only considers let's say what the customer said (that's assuming it would also identify when a extension received or made a call)?

HI Team

For now, we are sending the Transcribed text to ChatGPT and it is giving us the sentiment. So yes, if you are in doubt, it's okay I guess. ChatGPT can make mistakes. TIP 1: Promise me you will not fire anyone blindly for now because of the sentiment!! Check the recording first . Also your API key will start to learn more about your requests. So be patient.

When ChatGPT sometimes answers wrong, What do we do?
We somewhat accept it, and either live with it or we try again. So given AI happens to be my number 1 favourite subject let's do something interesting and try again - but this time maybe with a different prompt

3CX generates the Sentiment by getting the transcribed text, and sending this to openAI WITH a prompt. If we edit this prompt, or engineer it differently, you WILL get a different sentiment score - Maybe you can make it more strict!!

You can find this prompt in the custom parameter table - named: OPENAI_SENTIMENT

This means that you guys can experiment with this and prompt engineer your own prompt.

This is the prompt we crafted:

"Analyze the sentiment of the following call transcript on a scale from 1 to 5, where 1 is extremely negative, and 5 is extremely positive.
Consider professionalism, kindness, humor, empathy and satisfaction as positive traits and deduct points for rudeness, hostility, annoyance and unprofessionalism.
If the content is offensive, aggressive, vulgar, or inappropriate, provide an appropriate negative sentiment.
Only respond with an integer number from 1 to 5."

You want to change this from 1 to 10, Go ahead.
You want to make it more relaxed, strict, bureaucratic, politically correct? Go for it!!

Let us know how this goes.. Checking other questions as we speak.

 

[nikolascx]

Must I get to tier 1 for this to work? Currently in testing phase. Will it work with free tier?

HI Alphabetic
You need recording to be enabled. So without recording, nothing will work.
Everything in Audio AI requires audio.
Our version of Audio input is the recording file artefact. You need to have recording enabled first.
Thank you so much

 

[Alphabetic]

HI Alphabetic
You need recording to be enabled. So without recording, nothing will work.
Everything in Audio AI requires audio.
Our version of Audio input is the recording file artefact. You need to have recording enabled first.
Thank you so much

recording in 3CX?

i've got this on.

Had it set up for a few hours now and nothings happening.

Currently only on free tier.

 

[nikolascx]

Does anyone know how "OpenAPI - Whisper" compares to Google's transcription both in terms of price and transcription quality?

Hi CentrexJ
OpenAI is the absolute KING. No comparison. Unmatched
Price is also better but let's be generous and for educational purposes, leave the price out.
The Quality is out of this world. The auto detection of languages is stunning. The language support is Remarkable. Love it from start to finish. Just remember to put limits and price alarms.

 

[João Vitor]

HI Team

For now, we are sending the Transcribed text to ChatGPT and it is giving us the sentiment. So yes, if you are in doubt, it's okay I guess. ChatGPT can make mistakes. TIP 1: Promise me you will not fire anyone blindly for now because of the sentiment!! Check the recording first . Also your API key will start to learn more about your requests. So be patient.

When ChatGPT sometimes answers wrong, What do we do?
We somewhat accept it, and either live with it or we try again. So given AI happens to be my number 1 favourite subject let's do something interesting and try again - but this time maybe with a different prompt

3CX generates the Sentiment by getting the transcribed text, and sending this to openAI WITH a prompt. If we edit this prompt, or engineer it differently, you WILL get a different sentiment score - Maybe you can make it more strict!!

You can find this prompt in the custom parameter table - named: OPENAI_SENTIMENT

This means that you guys can experiment with this and prompt engineer your own prompt.

This is the prompt we crafted:

"Analyze the sentiment of the following call transcript on a scale from 1 to 5, where 1 is extremely negative, and 5 is extremely positive.
Consider professionalism, kindness, humor, empathy and satisfaction as positive traits and deduct points for rudeness, hostility, annoyance and unprofessionalism.
If the content is offensive, aggressive, vulgar, or inappropriate, provide an appropriate negative sentiment.
Only respond with an integer number from 1 to 5."

You want to change this from 1 to 10, Go ahead.
You want to make it more relaxed, strict, bureaucratic, politically correct? Go for it!!

Let us know how this goes.. Checking other questions as we speak.

Thank you for the response, the way sentiment analysis works it's crystal clear for us now!

 

[Evolute IT]

Hi CentrexJ
OpenAI is the absolute KING. No comparison. Unmatched
Price is also better but let's be generous and for educational purposes, leave the price out.
The Quality is out of this world. The auto detection of languages is stunning. The language support is Remarkable. Love it from start to finish. Just remember to put limits and price alarms.

Will the integration be able to detect the language? Right now we have to set one. We get calls in different languages so it would be useful

 

[JLowe]

In the Blog post it is mentioned that future editions of 3CX on Windows will be tied to enterprise. Does that mean we will have to have an Enterprise license to continue hosting on Windows?

Microsoft Windows Version of V20​

V20 Update 2 will also mark the first official release of the Windows version of Version 20. The release was delayed purely because of workload, there is no agenda to avoid companies using Windows as a server. However in future releases of 3CX the Windows version will be tied to the enterprise edition of 3CX. You can download the Windows version here.

 

[nikolascx]

Will the integration be able to detect the language? Right now we have to set one. We get calls in different languages so it would be useful

Hey @ConceptsWeb

This thing detects everything...It's SUPREMELY OUTSTANDING. Let me tell you a quick story,

One fine morning, after a night of research, we joined for a morning stand-up, 5 of us, in a product meeting.. Everything started business as usual until one of us (Leo) decided to go kamakazi on whisper and planned out the ultimate test to see whether it detects languages or not. (As openAI claim). The question on the table was - Will whisper model auto detect the spoken utterances?

(Actually we do happen to know a bunch of languages between us - so the scenario was like this)
1 meeting call, with 5 participants, and we went Acts of the Apostles style and started speaking in all the "tongues" we knew

So we started throwing in utterances in Maltese, Greek, French, Italian, German, Arabic, Cypriot (A rough and different Greek dialect), Spanish, Portuguese, Russian, Ukrainian and Leo threw in his Japanese also... coupled with different voices, tones, intonations et al. It was entertainment at it's best. We ended the call laughing our heads off. We have been in telephony for a long time now and together we had like 120 years of telephony experience. For sure we broke it.

Then we hanged up, extracted the audio from the video and transcribed it.

We were in total disbelief. We looked, and played the audio over and over again - because we could not believe. And we did this in an awkward but somewhat happy silence.

THIS THING IS ALIVE - It caught every single word we said and caught it correctly with total precision.

After that day, we never looked back. This one test gave us all the confidence we needed. We love it too much.. And we very much respect it also.

Have a great weekend to all!

 

[Alphabetic]

Hey @ConceptsWeb

This thing detects everything...It's SUPREMELY OUTSTANDING. Let me tell you a quick story,

One fine morning, after a night of research, we joined for a morning stand-up, 5 of us, in a product meeting.. Everything started business as usual until one of us (Leo) decided to go kamakazi on whisper and planned out the ultimate test to see whether it detects languages or not. (As openAI claim). The question on the table was - Will whisper model auto detect the spoken utterances?

(Actually we do happen to know a bunch of languages between us - so the scenario was like this)
1 meeting call, with 5 participants, and we went Acts of the Apostles style and started speaking in all the "tongues" we knew

So we started throwing in utterances in Maltese, Greek, French, Italian, German, Arabic, Cypriot (A rough and different Greek dialect), Spanish, Portuguese, Russian, Ukrainian and Leo threw in his Japanese also... coupled with different voices, tones, intonations et al. It was entertainment at it's best. We ended the call laughing our heads off. We have been in telephony for a long time now and together we had like 120 years of telephony experience. For sure we broke it.

Then we hanged up, extracted the audio from the video and transcribed it.

We were in total disbelief. We looked, and played the audio over and over again - because we could not believe. And we did this in an awkward but somewhat happy silence.

THIS THING IS ALIVE - It caught every single word we said and caught it correctly with total precision.

After that day, we never looked back. This one test gave us all the confidence we needed. We love it too much.. And we very much respect it also.

Have a great weekend to all!

Sounds like my weekends on the xbox in my party chat!

 

[Evolute IT]

Hey @ConceptsWeb

This thing detects everything...It's SUPREMELY OUTSTANDING. Let me tell you a quick story,

One fine morning, after a night of research, we joined for a morning stand-up, 5 of us, in a product meeting.. Everything started business as usual until one of us (Leo) decided to go kamakazi on whisper and planned out the ultimate test to see whether it detects languages or not. (As openAI claim). The question on the table was - Will whisper model auto detect the spoken utterances?

(Actually we do happen to know a bunch of languages between us - so the scenario was like this)
1 meeting call, with 5 participants, and we went Acts of the Apostles style and started speaking in all the "tongues" we knew

So we started throwing in utterances in Maltese, Greek, French, Italian, German, Arabic, Cypriot (A rough and different Greek dialect), Spanish, Portuguese, Russian, Ukrainian and Leo threw in his Japanese also... coupled with different voices, tones, intonations et al. It was entertainment at it's best. We ended the call laughing our heads off. We have been in telephony for a long time now and together we had like 120 years of telephony experience. For sure we broke it.

Then we hanged up, extracted the audio from the video and transcribed it.

We were in total disbelief. We looked, and played the audio over and over again - because we could not believe. And we did this in an awkward but somewhat happy silence.

THIS THING IS ALIVE - It caught every single word we said and caught it correctly with total precision.

After that day, we never looked back. This one test gave us all the confidence we needed. We love it too much.. And we very much respect it also.

Have a great weekend to all!

So can we set it without a language and it will detect automatically? Or is it required in 3CX to set the language like it was for Google?

 

[nikolascx]

So can we set it without a language and it will detect automatically? Or is it required in 3CX to set the language like it was for Google?

A yes - I forgot the main point - The language is not required - but the language is required for Google. We had to ship this fast.. So we had no time to rewrite/upset/retest the Google part.

We will tackle it later.

Anyway - it's so advanced, that even if you declare English and during convo it detects spanish utterance, it will give it to you no problem.

 

[N_G]

Hey @ConceptsWeb

This thing detects everything...It's SUPREMELY OUTSTANDING. Let me tell you a quick story,

One fine morning, after a night of research, we joined for a morning stand-up, 5 of us, in a product meeting.. Everything started business as usual until one of us (Leo) decided to go kamakazi on whisper and planned out the ultimate test to see whether it detects languages or not. (As openAI claim). The question on the table was - Will whisper model auto detect the spoken utterances?

(Actually we do happen to know a bunch of languages between us - so the scenario was like this)
1 meeting call, with 5 participants, and we went Acts of the Apostles style and started speaking in all the "tongues" we knew

So we started throwing in utterances in Maltese, Greek, French, Italian, German, Arabic, Cypriot (A rough and different Greek dialect), Spanish, Portuguese, Russian, Ukrainian and Leo threw in his Japanese also... coupled with different voices, tones, intonations et al. It was entertainment at it's best. We ended the call laughing our heads off. We have been in telephony for a long time now and together we had like 120 years of telephony experience. For sure we broke it.

Then we hanged up, extracted the audio from the video and transcribed it.

We were in total disbelief. We looked, and played the audio over and over again - because we could not believe. And we did this in an awkward but somewhat happy silence.

THIS THING IS ALIVE - It caught every single word we said and caught it correctly with total precision.

After that day, we never looked back. This one test gave us all the confidence we needed. We love it too much.. And we very much respect it also.

Have a great weekend to all!

That's a good story! I am sure OpenAI will be able to help 3CX in many areas even log analysis, reporting, call monitoring, answer suggestion and all that stuff.