How can we implement AI in our IVR system using Asterisk?

I want to implement AI in my IVR system to enable live voice conversations between the bot and callers, without using recorded messages.

This has been multiposted at How can we implement AI in our IVR system using Asterisk? - Development - FreePBX Community Forums although, I think you are confusing AI with the speech recognition application of it. For more general AI, there is far too little information to work on, unless all you want is to be able to include “AI” in your marketing brochures.

I think what they want is speech recgonition that’s fed in to an LLM that can then output to text to speech.

The LLM is the issue. They’re huge and system intensive. You won’t run this in your datacenter unless your datacenter is setup for crypto mining.

The technology is not there yet.

1 Like

I assume they will need good quality training data for that, as well.

On Monday 25 March 2024 at 16:56:47, david551 via Asterisk Community wrote:

I assume they will need good quality training data for that, as well.

Depends on what sort of service they want to provide to their callers :slight_smile:

Also, what sort of service this system is intended to replace…

Antony.

–
Was ist braun, liegt ins Gras, und raucht?
Ein Kaminchen…

                                               Please reply to the list;
                                                     please *don't* CC me.

For many cases you don’t need heavy LLM. Lightweight ASR + Lightweight TTS + small network for NLU is enough. Response time is within 0.5 second.

On Monday 25 March 2024 at 17:25:34, nshmyrev via Asterisk Community wrote:

For many cases you don’t need heavy LLM. Lightweight ASR + Lightweight TTS

  • small network for NLU is enough. Response time is within 0.5 second.

Do you have an example? Sounds interesting.

Antony.

–
Most people have more than the average number of legs.

                                               Please reply to the list;
                                                     please *don't* CC me.

Hi David:

I have been doing some development of natural language processing for Asterisk. Would be happy to share some of my insights with you.

Bill

Can we take the output of a call from our Asterisk system, convert it into text, and then pass it to an AI for processing? After the AI processes the text and generates an answer, can we convert that answer back into a voice format and pass it to the IVR in Asterisk to provide the response to the caller?It is possible?

The building blocks are in Asterisk to do so, yes. If what you actually want is an out of the box experience that requires no effort or development on your part, that is not in Asterisk.

We are currently working on this kind of technology. A pure agent.
As @dewdude has said LLMs can consume lots of resources, unaffordable for almost every single small/medium company even big companies. Thats why our approach is to make the technology small and distributed. As much it could be.

Our first release all the AI comes within our softphone. The transcription and summarisation (using a Phi-2 3B llm) happen in the phone itself and it’s delivered to your preferred backend via webhooks.

Yes, you’re right. I’m a PHP and Laravel developer, and I don’t have much knowledge or experience with Asterisk. My company wants me to work on it and create a system that can talk to humans automatically during calls. I’m learning the basics of Asterisk, but they want this feature, and I’ve already informed them that I don’t have experience in it, which is making it difficult for me. That’s why I keep posting in the community for things I can’t find on Google and youtube.

I have been working on a similar system (natural language processing). Happy to share what knowledge I have. Email me direct.

hi , we built voice bot using Rasa framework and we need to integrate with Asterisk. i am not finding any documents or video on this topic in google. so please help or guide me in integrating voice bot with asterisk

I usually integrate Asterisk with tiny bash o php script
Personally i use google-cli to convert wav to text, via recording caller speaking.(via bash script)
Not perfect but i advice to use unimrcp google dialogflow plugin fo AI integration. Dialogflow ES is a cheaper product than CX and for small purpouse is adequate.
For TTS instead i’m using amazon polly via php script