Talkative Voice AI Support

Welcome to the Talkative Voice AI support page. Here you’ll find everything you need to get started with our innovative Voice AI solution, including setup instructions, how it works, integration details, and troubleshooting tips.

Overview

Talkative Voice AI is an advanced, real-time voice assistant designed to handle incoming calls, capture conversation details, and even transfer calls when needed. Built with state-of-the-art speech-to-text (STT), large language models (LLM), and text-to-speech (TTS) technology, Voice AI delivers natural, conversational interactions for a range of use cases - from scheduling callbacks to confirming appointment details.

Voice AI bots can:

• Answer inbound calls automatically

• Capture key information using AI (like name, reference number, etc.)

• Lookup answers from your AI Knowledge Base

• Use custom flows and fallback logic to route customers to the right outcome

• Escalate to human agents when needed

Getting Started

Setting Up Your Voice AI

1. Create Your Voice Bot:

   As an Account Holder, simply add or edit a voice AI config, accessed in your menu. You will first need to ensure you have voice AI minutes (go to your billing and usage page) - if you do not have minutes, please reach out to support@gettalkative.com and we'll add some for you!

2. Create Your Voice AI Configuration:

   • Queue Label and Name: For internal purposes only
   • Voice: Choose your voice! If you can't find the one you want, please contact us, and we can add to the list
   • Max Duration: Specify the maximum duration (in seconds) for a call.
   • Introduction: Enter the greeting that your Voice AI will use (for example, “Hi, thanks for calling Talkative. How can I help you today?”).
   • Prompt: Provide the instructions for your Voice AI. See section below for more details.
   • Tools: Give your voice AI the ability to take actions, such as CRM lookups, or transferring the call. See section below for more details.

3. Configuration Outcome:

   Once set up, your configuration automatically creates a dedicated queue and user for handling voice interactions. A phone number is also provisioned.

Prompts

Unlike traditional decision-tree workflows, Voice AI is entirely driven by an LLM. This allows for highly flexible workflows and a superior customer experience, all accomplished with a simple set up. The LLM decides how to respond entirely based on the prompt and set of tools that you give it. Therefore, the prompt is the heart of the voice AI workflow - it dictates everything that happens in the call. 

New voice AI configs come with a preset prompt to help you get started. You can also view sample prompts here for inspiration: 

Talkative is actively working on a revised prompt build tool, estimated for release in Q4 2025, which will give users an even simpler way to build prompts and call experiences

The prompt should generally include the following, although this is a non-exhaustive list:

• Context: Explain to the AI (the LLM) its purpose and who it represents.

• Goals: Explain what you want it to do, for example it should aim to collect key information before transferring to the correct queue, or it should provide answers from the knowledge base.

• Conversational Paths: Instead of traditional decision-tree style workflows, you write these out in natural language. For example, if someone is calling up about a hotel booking, first check if it's a new or existing booking, then collect their name, then ask which resort they want, etc. This will likely form the bulk of the prompt. 

• Guidelines: This should generally not be tweaked too much from the default prompt setting, but you can add in guidance specific to your brand, perhaps telling it to not mention prices for example. This section is also where it should be explicit that the LLM must only use the info from the knowledge base rather than its own general knowledge.

• Style: This should generally not be tweaked too much from the default prompt setting - it guides the LLM to output messages in a format that is best suited for voice calls (e.g. keep messages short, date formats)

Talkative will soon be launching an automatic prompt grader. We're also always on hand to provide a review of your prompt - just get in touch. Here are some general tips and tricks:

• For best results, if you’re configuring Voice AI for a specific language, write the prompt in that language.

• Be careful using the same prompt for chat and voice, while in general they can be similar, remember that voice cannot do things like offer buttons, or present links.

• Avoid conflicting logic.
• Err on the side of being extremely explicit and clear.
• Examples are always useful.

Using Tools in Voice AI

Voice AI supports tools - structured actions that the AI can use during a conversation. These go beyond natural language responses, allowing your Voice AI to:

• Transfer the customer to another department or queue

• End the call

• Search your Knowledge Base directly

• Make external API calls with an HTTP request, e.g. booking a calendar appointment

These tools can be configured in the Voice AI bot builder under the Tools section. For full details on tool setup, prompts, and required parameters, see our configuration guide here.

How It Works

Talkative offers two approaches to voice AI:

1. Text-to-speech (TTS) combined with speech-to-text (STT)

• The default option for voice AI.

• Has the highest range of voices and language support.

• Offers the choice of underlying LLM, and therefore the best support for localised AI and data sovereignty.

• Works by:

  • 1. Voice Input & Transcription: Your caller’s voice is converted into text using a STT service.
  • 2. AI Processing: The transcribed text is sent to a large language model (LLM) that generates a contextually appropriate response based on your prompt.
  • 3. Response Delivery: The response text is converted into voice using a TTS service and played back to the caller in real time.

2. Speech-to-speech (S2S)

• A premium option for voice AI with higher costs.

• Is currently limited to OpenAI LLMs (US and EU options available).

• Is currently limited to American and British voices (although can speak multiple languages).

• Works by directly processing the incoming caller audio and responding with audio. By removing the steps to convert to and from text, it results in lower latency calls, and is better able to understand and convey tone/emotion.

Integration with your Phone System

Voice AI can be used to send calls into any phone system by sending in calls via PSTN or SIP (preferred). The call is received as an incoming phone call, and with SIP trunk integration you can keep your existing DID/DDI number for customers to dial. Please get in touch and we'll set up your SIP trunk.

Here’s how a typical call is handled:

1. Call Initiation:

   A caller dials your Voice AI number.

2. Interaction with Voice AI:

   • The Voice AI greets the caller.
   • The conversation is transcribed in real time.
   • Depending on the situation, the system either handles the call or initiates a transfer.

3. Call Transfer & Agent Support:

   When the conversation requires human intervention, the call is transferred (using SIP Refer) to your agent’s queue. Agents can view a screenpop showing a summary and full transcript of the conversation.

A real-time transcription of the call is also available to be displayed to the agent. To automatically present this to the agent, the below list of “advanced” phone system integrations are possible, with the list planned to expand over time:

• Mitel Integration: see our configuration guide here

• 8x8: planned, coming soon

• MS Teams: planned, coming soon

• Other phone systems: please get in touch and we'll make it happen!

Voice selection and language support

VoiceAI supports a wide range of languages and voices. You can choose from a variety of voice options within the bot editor, including different accents, genders, and tones.

The languages supported depend on the deployment model (S2S or TTS/STT), and vary further depending on the TTS/STT models.

For S2S, currently only English and American accents are supported, although the AI speaks multiple languages well.

Different combinations of STT/TTS providers allow different language/voice combinations.

TTS

• ElevenLabs is the recommended choice and is provided by default, with V3 supporting 74 languages: https://help.elevenlabs.io/hc/en-us/articles/13313366263441-What-languages-do-you-support

• Google provides a wider range of languages: https://cloud.google.com/text-to-speech/docs/list-voices-and-types#list_of_all_supported_languages

STT

• Deepgram is the recommended choice and is provided by default, with Nova-2-general supporting 30 languages: https://help.elevenlabs.io/hc/en-us/articles/13313366263441-What-languages-do-you-support

• Google provides a wider range of languages: https://cloud.google.com/speech-to-text/docs/speech-to-text-supported-languages 

Voice AI Playground & Phone Numbers

New voice AI configs come with an auto-generated US phone number. Please contact us if you need a number for a different region. 

You can also test your Voice AI bot directly in your browser using the Voice AI Playground. Just go to your Voice AI configuration and click “Test your Voice AI Bot”. This will launch an in-browser simulation that lets you try out your bot in real time - ideal for checking knowledge base lookups, call flows, and voice performance.

Note: If you have created a SIP trunk connection, the browser playground will not work, and you will need to dial the voice AI via your SIP infrastructure, not via the provided phone number.

Analytics & Reporting

All voice AI calls are transcribed in real-time. You can benefit from all of Talkative's analytics and reporting:

• View voice calls in Interaction logs, including AI call summaries: https://support.gettalkative.com/support/solutions/articles/201000057428-interaction-logs-overview

• Vice calls in Chat Monitoring https://support.gettalkative.com/support/solutions/articles/201000057435-realtime-supervisor-chat-video-monitoring

• View historical analytics: https://support.gettalkative.com/support/solutions/articles/201000057447-analytics-overview

• Analyse transcripts as part of AI Insights Reporting: https://support.gettalkative.com/support/solutions/articles/201000057438-ai-insights-analytics-reports

• Apply sentiment analysis: https://support.gettalkative.com/support/solutions/articles/201000057452-sentiment-analysis

• And more

You can also optionally turn on voice call recording at no extra cost. Recordings will show up in the Interaction Log page.

Voice AI Copilot

With an integration into a phone system (and soon with a universal agent phone connector), agents can view the transcript/summary of the call when it is transferred.

You can use AI data capture to collect key bits of information during the call: https://support.gettalkative.com/support/solutions/articles/201000083031-ai-interaction-data-capture

You can also allow the call to be transcribed in real-time when an agent is connected to the caller, allowing AI Agent Assist and real-time tools to support the agent with their call and wrap-up work.

You can view more here: https://support.gettalkative.com/support/solutions/articles/201000121143-voice-ai-copilot

Frequently Asked Questions (FAQs)

Q: What happens if my caller speaks in a language other than what’s configured?

A: While Voice AI can handle multiple languages, we recommend setting up one language per phone number to ensure the highest accuracy. If your callers use multiple languages, please contact support for guidance on configuration options. S2S allows calls to seamlessly transition between all languages mid-call. STT/TTS set up can transition between a limited number of languages, depending on the voice and STT/TTS settings used.

Q: Can I customize the voice and accent?

A: Yes! You can choose from a variety of voices and accents. We highly recommend testing different options to see which best represents your brand’s tone. With STT/TTS you can even create a cloned custom voice.

Q: How much does it cost?

A: Pricing is available on our websites for subscriptions. Please contact your Talkative CSM or channel partner if you wish to add voice AI to an existing subscription.