Voice AI Agents
A voice AI Agent is a Moveo AI Agent that talks to customers on a call. It uses the same guidelines, knowledge, tools and workflows as the AI Agent on chat and email. The voice layer adds speech recognition, turn detection and speech synthesis around it.
The voice pipeline
Every turn of a call goes through the same steps. The caller speaks, and the AI Agent replies in audio on the same call.
| Step | What it does | Where you set it |
|---|---|---|
| Noise suppression | Removes background noise from the caller audio. Off by default. | Voice integration, for example Telnyx Voice |
| Speech-to-text | Turns the caller audio into text. | Voice settings |
| Turn detection | Decides when the caller stops speaking. | Speech-to-text row of the voice settings |
| AI Agent and model | Picks the reply with the guidelines, knowledge and tools. | Voice channel of the model strategy |
| Text-to-speech | Turns the reply into audio in the AI Agent voice. | Voice settings |
To learn where the time of a turn goes, see Latency. To learn how the AI Agent handles pauses and interruptions, see Turn-taking.
Channels
| Channel | Inbound calls | Outbound calls |
|---|---|---|
| Twilio Voice | Yes | Yes, with campaigns |
| Telnyx Voice | Yes | Yes, with campaigns |
| Web Voice | Yes | No |
- Twilio Voice and Telnyx Voice connect your own Twilio or Telnyx account. Customers call your phone number, or the AI Agent calls them from it.
- Web Voice adds a voice button to the Moveo web widget. Visitors start a call from the browser. To turn it on, select a Web Voice integration in Voice mode of the web widget.
- For outbound calls, see Outbound calls and Answering machine detection.
Where you set what
A voice AI Agent has three layers of settings. Each layer has one job.
| Layer | Holds | Page |
|---|---|---|
| Integration | Telephony: phone number, call recording, AI Agent starts first, noise suppression, answering machine detection, SIP caller, and the allowed origins of Web Voice. | Twilio Voice, Telnyx Voice, Web Voice |
| AI Agent | Speech-to-text, turn detection and text-to-speech for each language, the language model of the Voice channel, and the starting dialog. | Voice settings, Model strategy |
| Account | Your own keys for speech providers and language model providers. | Speech models, Language models |
You set speech-to-text, text-to-speech and turn detection once, in the voice settings of the AI Agent. Every phone line and web widget that connects to the AI Agent then uses the same settings.
What Moveo adds around the call
A voice framework gives you the pipeline. Moveo also gives you the customer-service operation around the call.
| Capability | What you get |
|---|---|
| Human handover | The Handover action transfers the call to a phone number or a SIP URI. See Call transfer. |
| Inbox | Human agents answer the chat and email conversations of the same AI Agent in the inbox. |
| Analytics and logs | Calls show in analytics. The AI Agent logs hold the transcript and the call recording. |
| Campaigns | Campaigns call a contact list with the AI Agent. |
| Simulations | Simulations run regression tests with the Voice channel type. |
| Versions | You publish a version and route calls to it with a rule. |
| Regions | Your account runs in Europe, the United States or Brazil. See Outbound IP addresses. |
Web Voice calls do not support handover. Gate the Handover
action in your workflow so it does not run on browser calls.
Voice documentation map
| Page | Read it to |
|---|---|
| Choosing a voice AI platform | Compare Moveo with a voice framework or a voice developer API. |
| Path to production | Go from a new AI Agent to live calls, step by step. |
| Voice settings | Select the speech-to-text and text-to-speech providers. |
| Speech models | Connect your own speech provider keys. |
| Languages | Set up an AI Agent that speaks more than one language. |
| Voice guidelines | Write guidelines that sound natural on a call. |
| Pronunciation | Fix how the AI Agent says names, numbers and terms. |
| Latency | Learn where the time of a turn goes and how to lower it. |
| Turn-taking | Tune turn detection and interruptions. |
| Outbound calls | Call customers with campaigns. |
| Answering machine detection | Detect voicemail on outbound calls. |
| Call recording | Store and play back calls. |
| Test voice agents | Test with Web Voice calls, phone calls and simulations. |
| Security and compliance | Review data handling and compliance. |
| Call transfer | Transfer a call with Standard transfer or SIP REFER. |