This guide aims to help you understand the integration points for the Voice AI system. This guide assumes you are comfortable using SIP and setting up a SIP Trunk in Mitel. Each section contains a dropdown with a link indicating the minimum known support version and a setup guide. Additionally, there is information to support decisions around licensing relating to SIP channels required for operation.

To find your SIP trunk details, navigate to a Voice Route config page in engage and you will be able to see the SIP Trunk which has been configured for you:

This will be displayed at the top of each Voice Route for ease of access.

Talkative utilises Twilio for SIP Trunk application capabilities. Twilio’s resilient managed trunks ensure stability when routing Voice AI calls. To ensure the signalling and RTP traffic can be received, please whitelist the signalling IPs which can be found here: IP Addresses for Elastic SIP Trunking Services | Twilio - The default region is none selected is us1. Each region has 4 IP addresses which signalling traffic may originate from or be routed too. We recommend always using the FQDN and never a direct IP address.

By default the SIP Trunk region is us1 - if you wish to use a different region, you may adjust your Trunk FQDN to include a regional identifier:

This will force your signalling to favour a particular region for originating it’s traffic.

The most common setups include a Mitel Border Gateway - this will contain your at edge SIP Trunk configuration. Other configuration may include a custom or bespoke SBC. This can be used in place of an MBG. Whilst an SBC is commonly found, it is not a pre-requisite, and a PBX can be integrated with directly.

The current supported PBX call handler systems include MiVB, MX-One and MV-5000

Whilst not required, you can additional use contact centre platforms to allow for screen pops within platforms such as Web Ignite - this additionally allows for IVR workflows to add contextual metadata to your interactions such as the console URL and any passed interaction data.

The amount of SIP channels you consume will depend on if you are utilising Voice Assist in your workflows. The details for the channels are as follows:

Voice AI without Voice Assist will consume a maximum of 2 SIP channels. The usage is defined as follows:

PSTN/CC ↔ PBX: 1 SIP/media leg on their side.

PBX ↔ Talkative VAI trunk: 1 SIP/media leg on their side.

Voice Assist, either direct or including Voice AI triage will consume a maximum of 3 SIP channels. The usage is defined as follows:

PSTN/CC ↔ PBX: 1 leg.

PBX ↔ Talkative VAI trunk (initial call to AI): 1 leg.

Talkative VAI ↔ PBX (AI dials back / second leg to stay in): 1 leg.

The additional channel is consumed as the system makes a SIP Dial instead of a SIP Refer to remain in the call, essentially creating a 3 way call bridge.

Voice AI minutes have three different consumption multipliers. They are:

Minutes are consumed in bundles and then the rate of consumption depends on the tool utilising it. These minutes are then allocated within Engage according to the required breakdown.

An example of this might be a user who requires 1000 Standard Voice AI minutes per month and 5000 Voice Assist Minutes.

Another example might be a customer wishing to use premium speech to speech models. If they required 1000 minutes of Premium Voice AI, and 5000 minutes of Voice Assist, the breakdown would be as follows:

Token usage should be considered alongside SIP trunk consumption, as the amount of minutes available heavily determines the likelihood of total call concurrency and as such, the amount of concurrent sip legs which will be established.

The calculation for Standard and Premium voice is very similar and should be determined by working out total expected call volume and then an estimating the time a customer is likely to spend talking to the bot. The length of time they will spend with the bot will be determined by the bots abilities. For example, a simple bot which just collects a name and account number and then passed the call to a human agent may have an expected duration of 60 seconds. Conversely, a complex bot which will collect data and use tools to interact with an API or utilising KB lookups, provide customer information may last for 5-10 minutes.

If you have multiple bots covering different scenarios, ie. a triage bot which acts as an IVR and a more complex bot to handle CRM interactions, you may find benefit to calculating their expected usage separately and then combine the values when you purchase.