> For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.ada.cx/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.ada.cx/_mcp/server. # Voice onboarding ## Overview Onboard your AI Agent to handle Voice conversations over the phone by following the steps in this guide. ## Limitations Voice onboarding has the following constraints: * **Feature availability**: This feature may not be included with your organization's subscription package. For more information, contact your Ada team. ## Use cases A Voice AI Agent allows you to engage with end users over the phone. * **Automate phone support**: Enable your AI Agent to answer calls and assist end users without human intervention. * **Support multiple languages**: Configure your AI Agent to speak with end users in their preferred language, with regional dialect options. * **Brand your AI Agent's voice**: Select a speaking voice that matches your organization's identity. * **Improve recognition of specialized terms**: Flag brand names, product names, or industry jargon for Voice detection in Glossary, so the speech-to-text engine transcribes them accurately. ## Capabilities & configuration Your Voice AI Agent includes the following capabilities: * **Twilio integration**: Connect using Ada-managed Twilio (recommended) or your own Twilio account with phone numbers, SIP domains, or Twilio Applications * **Multilingual support**: Enable Voice for any of 42 supported languages, with regional dialect options for select languages * **Language selection menu**: Let callers choose their preferred language at the start of a call * **Speaking voice**: Select an AI Agent voice and speaking pace for each enabled language * **Voice detection**: Flag brand names, acronyms, or jargon in Glossary so speech-to-text recognizes them * **Agent SMS**: Control whether your AI Agent can text callers during a call to capture information, send instructions, or deliver a CSAT survey link * **Your own SMS channel**: Route in-call texts through your own SMS provider and phone number ## Quick start Get your Voice AI Agent up and running in a few steps. For detailed instructions, see [Implementation & usage](#implementation--usage). Work with your Ada team to provision Twilio resources (phone number, SIP domain, or Twilio Application). Most customers use Ada-managed Twilio. On the Ada dashboard, go to **Config > CHANNELS > Voice** and configure your Twilio settings if using bring-your-own Twilio. Enable [Languages](/docs/setup/languages) for Voice and choose regional dialects on the **Dialects** tab. Select a speaking voice for each enabled language on the **Speaking voice** tab. Configure caller silence timeout. To improve recognition of brand and product names, flag those terms for Voice detection under **Config > AI Agent > Glossary**. Test your AI Agent over voice from the dashboard using **Test AI Agent > Voice**. ## Implementation & usage Configure your Voice AI Agent by completing the following steps in order. ### Twilio and your Voice AI Agent Ada’s Voice AI Agent is powered by **Twilio**. In almost every deployment, Ada provisions and manages the necessary Twilio resources for you—no separate Twilio contract or configuration required. #### Ada-managed (recommended) Most Voice customers use Ada’s central Twilio account. * An Ada team member will provision and configure one or more of the following so your AI Agent can receive calls and send and receive SMS: * **Twilio phone number** * **Twilio SIP domain** * **Twilio Application** * Ada maintains the Twilio configuration, monitors usage, and absorbs the telephony costs. * When this setup is complete, go straight to **[Enable Voice in Multiple Languages](#multilingual)** to continue onboarding. #### Bring-your-own Twilio account (optional) If you must use your own Twilio project—for regional compliance or existing contracts—expand the steps below to prepare your: * **Twilio phone number** * **Twilio SIP domain** * **Twilio Application** ### Set up your own Twilio resources #### Set up your own Twilio resources 1. Set up one or more phone numbers with Twilio, using one of the following options: * Purchase a phone number through Twilio. For more information, see [How to Search for and Buy a Twilio Phone Number from Console](https://support.twilio.com/hc/en-us/articles/223135247-How-to-Search-for-and-Buy-a-Twilio-Phone-Number-from-Console) at Twilio's Help Center. * Port an existing phone number from another service provider to Twilio. For more information, see [Porting a Phone Number to Twilio](https://support.twilio.com/hc/en-us/articles/223179348-Porting-a-Phone-Number-to-Twilio) at Twilio's Help Center. 3. Submit your organization's toll-free phone number(s) to Twilio for verification. * If you are using a toll-free number with your AI Agent and you want to be able to send SMS messages from it during a phone conversation, you will need to submit a Toll-Free Message Verification request to Twilio. For more information, see [Toll-Free Message Verification for US/Canada](https://support.twilio.com/hc/en-us/articles/5377174717595-Toll-Free-Message-Verification-for-US-Canada) at Twilio's Help Center. * If you are using a local number with your AI Agent and you want to be able to send SMS messages from it during a phone conversation, you will need to submit an A2P (Application to Person) 10DLC (10 digit long code) registration request to Twilio. For more information, see [Programmable Messaging and A2P 10DLC](https://www.twilio.com/docs/messaging/compliance/a2p-10dlc) at Twilio's Help Center. 5. (Optional) Set up a SIP domain with Twilio. Your AI Agent can also receive calls from your contact center infrastructure, transferred over the public internet using Session Initiation Protocol (SIP). To do this, you first need to create a SIP domain in Twilio. 1. Add at least one IP Access Control List (ACL) or Credential List in Twilio An Access Control List (ACL) is a security measure that manages who can access your network resources by specifying allowed IP addresses. In SIP communications, using an ACL enhances security by ensuring only trusted traffic can reach your Twilio SIP domain, preventing unauthorized access. This procedure takes you through configuring an IP ACL in Twilio, so Twilio only accepts SIP traffic that originates from your specified IPs. You can also choose to configure credential lists instead, so it only accepts traffic from specific usernames and passwords. For more information, see [Inbound - Sending SIP to Twilio](https://www.twilio.com/docs/voice/api/sending-sip#ip-access-control-lists-acl) at Twilio’s Help Center. 1. Log into [Twilio Console](https://www.twilio.com/login). 2. In your Twilio account, go to **Develop** > **Voice** > **Manage** \> **IP access control lists**, then click **Create new Access Control List**. The Create new Access Control List window opens. 3. In the Create new Access Control List window, fill in the following fields: 1. Under **ACL Friendly Name**, enter a descriptive name that represents the purpose of the list (e.g., `My Contact Center Platform IP Address`). 2. Under **IP Address Range Friendly Name**, enter a name to help identify the IP Address range. 3. Under **CIDR Network Address** and **Range**, enter the specific IP address range that should be allowed access. 5. Click **Create new ACL** to save the list and close the window. Then, click **Create new IP Address Range** to add an additional IP range to the list. The Add IP Address Range window opens. 6. Repeat the previous steps for adding an IP Address Range. 7. Click **Save**. 5. Create a new SIP domain. 1. Go to **Develop** > **Voice** > **Manage** > **SIP domains**. 2. Click the plus sign to create a new SIP domain. 3. Create a SIP URI/domain name. This property allows you to specify a global unique SIP domain that is used to route SIP traffic from your infrastructure over the public internet to the correct server hosted by Twilio in the cloud. For more information on SIP domains, see [SIP URI/domain names](https://www.twilio.com/docs/voice/api/sending-sip#sip-uri--domain-name) at Twilio’s Help Center. 4. If you wish to manually connect to a specific geographic Edge Location that is closest to the location of your communications infrastructure, see [Localized SIP URIs](https://www.twilio.com/docs/voice/api/sending-sip#localized-sip-uris) at Twilio’s Help Center. This can reduce latency and improve call audio quality. 5. Under **Voice Authentication**, beside **IP Access Control Lists**, select both of the IP Access Control Lists you created. 6. Under **Secure Media**, click the toggle so it says **Enabled**. 7. Click **Save**. After getting your phone number and/or SIP domain set up with Twilio, you can connect it with your AI Agent. ## Connect your Twilio account with your Ada AI Agent In your Twilio account, you need to obtain your account string identifier (SID), and create an API key to use with Ada. 1. On the Ada dashboard, go to **Config > CHANNELS > Voice**, then go to the **Configuration** tab. Keep this page open as you go into your Twilio account to get your account details, so you can paste them in as you find them. 2. In your Twilio console, in your account info, find your Account SID. Your Account SID is a unique identifier that acts like a username when you're using Twilio's API. For information on how to find it, see [Auth Tokens and How to Change Them](https://support.twilio.com/hc/en-us/articles/223136027-Auth-Tokens-and-How-to-Change-Them) at Twilio's Help Center. 3. On the Ada dashboard, under **Twilio Configuration**, paste your Account SID into the **Twilio Account SID** field. 4. In your Twilio console, create a new API key. Enter a name into the **Friendly name** field, set the **Key Type** to **Main**, and leave the other settings with their default values. For information on how to create an API key, see [API Keys and How to Change Them](https://support.twilio.com/hc/en-us/articles/9318455807771-API-Keys-and-How-to-Change-Them) at Twilio's Help Center. When you create an API key, you get both a key and a secret. You can only view the secret once for security reasons, so make sure you save it in a safe place. 6. On the Ada dashboard, under **Twilio Configuration**, paste the API key into the **Twilio API Key SID** field, and paste the secret into the **Twilio API Key Secret** field. 7. Configure your AI Agent to use a phone number, a SIP domain, or a Twilio Application to receive phone calls. Your AI Agent can only use one of these methods to receive phone calls. Click a section to expand the full instructions. For more information on how you can configure your AI Agent with your contact center platform, see [Integrate your voice AI Agent with your Contact Center as a Service (CCaaS) provider](/docs/channels/voice/contact-center-integration). ### Configure your AI Agent to receive phone calls using a phone number #### Configure your AI Agent to receive phone calls using a phone number 1. Select **Phone Number**. 2. Under **Default phone number**, enter your Twilio phone number. This is the phone number your AI Agent will use to receive phone calls. 3. If required, you can add additional numbers if you want to use them to serve different audiences (e.g,. end users who want to speak in specific languages, or if your AI Agent serves multiple brands). To do this, click **Add another phone number**, add the number, and select a default language for it. 4. Under **Fallback phone number**, enter a fallback number. Ada will route calls to this phone number in the case of system downtime that prevents your AI Agent from responding. 5. If required, select the **Use a different phone number for SMS** check box. Under **Phone number for SMS and test conversations**, enter the phone number. Your AI Agent sends and receives SMS messages from this number instead of your main phone number. The check box appears only while [Agent SMS](#control-whether-your-ai-agent-can-send-sms) is enabled. 6. Click **Save**. ### Configure your AI Agent to receive phone calls using a SIP domain #### Configure your AI Agent to receive phone calls using a SIP domain 1. Select **SIP domain**. 2. Enter a SIP domain. This should follow the format `your-domain-name.sip.twilio.com` and should be the global unique SIP domain that you previously created. If you haven’t created a SIP domain yet, entering and saving a SIP domain here will create the domain in your Twilio account. You will then need to configure at least one IP Access Control List or Credential List in Twilio to use with this SIP domain. 4. If required, you can add additional SIP domains that you created in Twilio if you want to use them to serve different audiences (e.g,. end users who want to speak in specific languages, or if your AI Agent serves multiple brands). To do this, click **Add another SIP domain**, add the domain, and select a default language for it. 5. Under **Fallback SIP address**, enter a fallback SIP address. Ada will route calls to this SIP address in the case of system downtime that prevents your AI Agent from responding. 6. For each SIP domain, under **Phone number for SMS**, enter a Twilio phone number. Your AI Agent sends and receives SMS messages from this number for calls on that domain. The field is optional and appears only while [Agent SMS](#control-whether-your-ai-agent-can-send-sms) is enabled. 7. Under **Test phone number**, enter the Twilio phone number Ada uses for outbound test calls. While Agent SMS is enabled, your AI Agent also uses this number for SMS on any SIP domain that has no **Phone number for SMS**. 8. Click **Save**. ### Configure your AI Agent to receive phone calls using a Twilio Application #### Configure your AI Agent to receive phone calls using a Twilio Application 1. Select **Twilio Application**. When you select this option, Ada prepares to create a new Ada Twilio Application which it uses to receive phone calls. 2. Enter the **phone numbers used to receive calls.** * Under **Phone numbers used to receive calls**, enter each phone number that your end users might dial before the call is transferred to your Ada Twilio Application. * For each phone number that is used to receive calls, select the default language for it. When Ada receives a call that was made to that phone number, the AI Agent will default to the configured language. * If there are multiple different phone numbers that end users dial before the call is transferred to Ada (e.g. different phone numbers for different audiences, languages, or brands), click **Add another phone number**, enter the phone number, and select a default language for it. * **Note**: If Ada's Twilio Application receives an inbound call on a phone number that isn’t listed here, the AI Agent defaults to English. 4. Under **Fallback phone number**, enter a fallback phone number. Ada will route calls to this phone number if, for any reason, your AI Agent becomes unresponsive and cannot answer calls. 5. Under **Test phone number**, enter a phone number from your Twilio account. Ada uses this number for outbound test calls. While [Agent SMS](#control-whether-your-ai-agent-can-send-sms) is enabled, your AI Agent also uses it to send and receive SMS messages. 6. Click **Save**. * Ada creates the Twilio Application in your Twilio account. * Once saved, the **Twilio Application ID** field appears. This is your Twilio App SID. Copy this value for use in your phone system or contact center platform. You can configure additional settings to fine-tune your AI Agent's behavior during phone calls. Additional settings are covered in later topics. > **Note** > > Your usage of Voice is subject to both applicable privacy and phone > communication laws, and Twilio's policies. These may vary based on your > location and industry. If you have any questions, please contact Ada. ### Enable Voice in multiple languages\[#multilingual] Your AI Agent can be enabled to speak any of Ada's Voice-supported languages over the phone. For the full list, see the [Supported languages table](/docs/setup/languages/about-multilingual-support#supported-languages) in About multilingual support. When you enable a language in your AI Agent, it is available across all channels that support the language. You cannot enable a language for one channel only. #### Choose regional dialects for your languages\[#multilingual-dialects] Some of the languages available in Voice also have support for different regional dialects of those languages: * **Arabic** * Egypt * Kuwait * Lebanon * Morocco * Saudi Arabia * United Arab Emirates * **Chinese (Traditional)** * Hong Kong * Taiwan * **Dutch** * Belgium * Netherlands * **English** * Australia * Canada * Jamaica * Philippines * Singapore * United Kingdom * United States * **French** * Canada * France * **Portuguese** * Brazil * Portugal * **Spanish** * Mexico * Spain * United States The dialect you choose narrows down which speaking voices will be available for that language. When you choose a dialect for your AI Agent, your AI Agent will automatically use the default speaking voice for that language and dialect. You can always go into your settings and [choose a different voice](#UUID-e17eb05a-42bb-6a3d-e842-36988f98b48c_section-idm4503712244217633647134198894) for that dialect instead. **To change the dialects available for your AI Agent:** 1. On the Ada dashboard, go to **Config > CHANNELS > Voice**, then go to the **Dialects** tab. 2. For each language that has dialects available, in the **Dialect/region** list, select an available dialect. 3. Click **Save**. Your AI Agent immediately starts using the dialects you chose for all conversations in the corresponding languages, for both internal testing and for any live voice content in your AI Agent. #### Understand how your AI Agent determines the end user's language\[#multilingual-end-user-language] By default, your AI Agent determines the language to use with your end user based on the `language` metavariable. There are a few ways you can set the value of this metavariable: * **Set a default language for the phone number or SIP domain** For each phone number or SIP domain that your AI Agent uses to receive phone calls, you will need to assign a default language. When a call is received using that phone number or SIP domain, the end user's `language` metavariable will automatically be set to that language. * **Let end users select from a menu of enabled languages** You can begin all of your AI Agent's calls with a menu that lets end users choose one of your enabled languages. The Agent plays the menu before the [Greeting](/docs/automation/greetings), even when the number is associated with a specific language. You can also present a different menu depending on the end user's context. For example, you can offer one set of languages for a number dialed in Canada and another for a number dialed in the United States. 1. On the Ada dashboard, go to **Config > CHANNELS > Voice**, then go to the **Language selection** tab. 2. Turn on the **Language selection menu** toggle. 3. In the **Default menu**, add at least two languages with **Add language**. This menu plays whenever no conditional menu matches. Each language is assigned a number, in list order, that the end user dials on their keypad to select it. Drag languages to change the order. 4. To vary the menu by context, select **Add menu** and set up a conditional menu. Under **Show this menu when**, add one or more conditions based on end user or call information, such as the number that was dialed, then add at least two languages. 5. Conditional menus are evaluated from top to bottom. The first menu whose conditions match is played; otherwise the **Default menu** is used. Use the up and down arrows to change a menu's priority. * **Use the End Users API** You can use the End Users API to [update a user's profile](../../../reference/end-users/patch-end-user-by-id) to set the `language` metavariable. * **Send language information in the SIP header** If you're using a SIP domain for your AI Agent to receive phone calls, you can set the conversation language by including a language value in one of the following SIP headers. This overrides the default language configured for that SIP domain. The language value must be a [BCP-47 language code](https://en.wikipedia.org/wiki/IETF_language_tag) that matches a dialect enabled on your Voice configuration page (for example, `es-ES` or `fr-CA`). There are two supported methods, listed in priority order: 1. **User-to-User (UUI) header** — Include a `language` key-value pair in the SIP `User-to-User` header. For example: `User-to-User: language=es-ES`. 2. **SIP X-header** — Send the language value in an `X-language` header. For example: `X-language: es-ES`. If both headers are present, the UUI header value takes precedence. * **Send language information as a Twilio Parameter** If you're using a Twilio Application for your AI Agent to receive phone calls, you can include the language value as a parameter when transferring a call to the Application. ### Choose a speaking voice for your AI Agent\[#UUID-e17eb05a-42bb-6a3d-e842-36988f98b48c\_section-idm4503712244217633647134198894] Choosing a speaking voice for your AI Agent is an important part of your AI Agent's branding. You can change your AI Agent's voice at any time to better fit your AI Agent's branding needs. **To choose a speaking voice:** 1. On the Ada dashboard, go to **Config > CHANNELS > Voice**, then go to the **Speaking voice** tab. 2. Under **Speaking voice**, select a voice for each language you have enabled in your AI Agent. 3. Optionally, in the text field, you can type in a text sample, like a message from your AI Agent, so you can hear your AI Agent read out a specific piece of your AI Agent content. Then, you can click the **Play** button beside a selected voice to play the sample. 4. Click **Save**. Your AI Agent immediately starts using the voices you selected for both internal testing and for any live voice content in your AI Agent. #### Adjust how fast your AI Agent speaks You can set a speaking pace for each language your AI Agent speaks. A slower pace can make speech easier to follow for audiences like older callers or non-native speakers, while a faster pace suits short, transactional calls. **To adjust the speaking pace:** 1. On the Ada dashboard, go to **Config > CHANNELS > Voice**, then go to the **Speaking voice** tab. 2. In each language's row, use the **Pace** slider to choose one of four settings: **Slowest**, **Slow**, **Normal**, or **Fast**. The default is **Normal**. 3. Optionally, click the **Play** button beside the selected voice to hear a preview at the selected pace. You can type your own text sample into the text field to hear how specific content will sound. 4. Click **Save**. Your AI Agent immediately starts speaking at the pace you selected for both internal testing and for any live voice content in your AI Agent. #### How read-backs affect the speaking pace When your AI Agent reads a captured value back to a caller to confirm it — an order number, an email address, a date — it speaks that turn at the **Slow** pace, then returns to the pace you selected for the rest of the call. This happens automatically; there is nothing to turn on. The read-back pace only ever slows speech down, never speeds it up: * If you selected **Slowest**, read-backs stay at **Slowest**. * If you selected **Slow**, read-backs are unchanged. * If you selected **Normal** or **Fast**, read-backs drop to **Slow**. This applies to read-backs in [Playbooks](/docs/automation/playbooks) Ask steps and to [Actions](/docs/automation/tools/api-tools) inputs. Whether your AI Agent reads a value back at all is controlled by the [Voice confirmation settings](/docs/channels/voice/voice-configuration/voice-call-capture-options#voice-confirmation-settings) on each input or Ask-step variable. #### When callers ask your AI Agent to change its pace Callers can ask your AI Agent to speak faster or more slowly during a call. Your AI Agent adjusts how it speaks for the rest of that call. How far the pace moves depends on what the caller says: * **A direct request to slow down moves two settings.** "Slow down" or "you're talking too fast" drops the pace two steps. An AI Agent set to **Normal** moves to **Slowest**. * **An indirect signal moves one setting.** Your AI Agent also slows down when a caller says something like "I can't understand you". That moves one step, because the caller did not ask about speed directly. * **A request to speed up moves one setting.** This applies however directly the caller asks. These rules apply to every request: * The pace moves along **Slowest**, **Slow**, **Normal**, and **Fast**. At either end of the range, further requests in that direction change nothing. An AI Agent already speaking at **Fast** stays at **Fast** if asked to speed up. * The change applies to that call only. The **Pace** you selected on the **Speaking voice** tab is not modified. The next call starts from it again. * Only the delivery changes. Your AI Agent says the same thing. It stays on the same step of its [Playbook](/docs/automation/playbooks). Read-backs use whichever is slower, the read-back pace or the pace the caller settled on. A caller who sped the AI Agent up still hears captured values read back more slowly. A caller who slowed it to **Slow** or **Slowest** hears read-backs at that same pace, because the call is already at or below the read-back pace. ### Configure how long an end user can be silent before your AI Agent hangs up\[#UUID-64669525-022b-0e3b-1569-53881337d173\_section-idm43365062568912] You can choose how much time your AI Agent will wait for an end user to speak before it reprompts the end user. **To configure maximum caller silence:** 1. On the Ada dashboard, go to **Config > CHANNELS > Voice**, then go to the **Configuration** tab. 2. Under **Maximum caller silence (seconds)**, enter the number of seconds you want your AI Agent to wait before re-prompting the end user. After re-prompting the end user, the AI Agent will wait an additional 10 seconds before then ending the call. 3. Click **Save**. ### Control whether your AI Agent can send SMS Your Voice AI Agent can text an end user during a call. It can [capture information](/docs/channels/voice/voice-configuration/voice-call-capture-options) that is hard to say aloud, [send instructions or links](/docs/channels/voice/voice-configuration/voice-tools#voice-send-sms), or deliver a [CSAT survey link](/docs/channels/voice/voice-configuration/voice-csat-configuration) when the call ends. The **Allow AI Agent to send and receive SMS** toggle controls whether your AI Agent can start any of these texts. This guide calls the toggle Agent SMS. The toggle is **enabled by default** for all Voice AI Agents, so nothing changes until you disable it. While Agent SMS is disabled: * Your AI Agent never offers to text an end user and never asks for [SMS consent](/docs/channels/voice/voice-configuration/sms-consent). * The [Send SMS tool](/docs/channels/voice/voice-configuration/voice-tools#voice-send-sms) is not available to your AI Agent. * **Speech and SMS** is removed from the [Voice call capture options](/docs/channels/voice/voice-configuration/voice-call-capture-options). An input already saved as **Speech and SMS** is captured by speech only. * [CSAT for Voice](/docs/channels/voice/voice-configuration/voice-csat-configuration) is unavailable, because the survey link is delivered only by SMS. * For a phone number, the **Use a different phone number for SMS** check box is hidden, together with the **Phone number for SMS and test conversations** field. For a SIP domain, the **Phone number for SMS** field is hidden. The saved numbers are not cleared, and the fields appear again when you enable Agent SMS. Agent SMS stops the texts that your AI Agent starts during a call or at the end of one. **To control whether your AI Agent can send SMS:** 1. On the Ada dashboard, go to **Config > CHANNELS > Voice**, then go to the **Configuration** tab. 2. Under **Allow AI Agent to send and receive SMS**, enable or disable the toggle. 3. Click **Save**. ### Send texts from your own SMS channel By default, the text messages your AI Agent sends during a call come from the Twilio number on this page. You can send them from your own SMS provider and phone number instead. This needs a developer. They create an SMS channel with the Conversations API and run a relay, a small service that receives each text message from Ada and sends it through your provider. When the relay runs, you turn on the **Use your own SMS channel** toggle here. For the developer steps, see [Relaying voice texts through your own SMS provider](/reference/conversations/developer-guides/relaying-voice-texts-through-your-own-sms-provider). Before you choose this, know what changes: * Ada cannot confirm that a text message was delivered. Your provider has that record. * A caller who replies STOP reaches your provider, not Ada. Your team handles opt-outs. * Carrier registration and compliance for your number are yours. The toggle is **off by default**. It appears only while Agent SMS is enabled. While your AI Agent has no SMS channel, you cannot turn the toggle on. The page then shows **Set up an SMS channel with the Conversations API to turn this on.** Before you turn on the toggle, subscribe an enabled [webhook endpoint](/reference/webhooks/overview) to `v1.conversation.message`. The dashboard does not check your webhook endpoints. While no enabled endpoint subscribes to that event, Ada drops each text message. Ada does not send it from the Twilio number instead. While the toggle is on: * Each text message the AI Agent sends during a call goes to your relay as a `v1.conversation.message` webhook. Your relay sends it with your provider. * The caller can reply by text. Your relay posts the reply to the Conversations API, and the AI Agent answers by voice on the call. * The CSAT survey link the AI Agent sends when it ends the call also goes to your relay. * The **Use a different phone number for SMS** and **Phone number for SMS** fields stay on this page. The AI Agent does not use those numbers for text messages during a call. What the caller experiences: * When the AI Agent texts to collect an answer, it waits for the reply. In a Playbook Ask step or an Action input, the caller can still answer by voice. For an input captured in a Process, the AI Agent asks the caller to reply to the text message. * If the caller stays silent, the AI Agent ends the call after the [caller silence limit](#UUID-64669525-022b-0e3b-1569-53881337d173_section-idm43365062568912). That limit is also the time your relay has to deliver the text message and post the reply. * When the AI Agent texts instructions or the CSAT survey link, the call continues or ends as usual. The AI Agent does not wait. Turn the toggle off to send text messages from the Twilio number again. A relay that still posts replies then receives an error. A change to the toggle takes effect within a minute, and it can apply to a call that is in progress. **To send texts from your own SMS channel:** 1. On the Ada dashboard, go to **Config > CHANNELS > Voice**, then go to the **Configuration** tab. 2. Under **Use your own SMS channel**, turn on the toggle. 3. Click **Save**. ### Configure background noise cancellation\[#background-noise-cancellation] Your Voice AI Agent can filter out background noise — such as traffic, crowds, background conversations, and other ambient sounds — before it reaches speech detection and transcription. This prevents background sounds from being mistranscribed as end user speech, which can interrupt conversational flow and reduce your AI Agent's ability to accurately understand end users. Background noise cancellation is **enabled by default** for all Voice AI Agents. **To enable or disable background noise cancellation:** 1. On the Ada dashboard, go to **Config > CHANNELS > Voice**, then go to the **Configuration** tab. 2. Under **Background noise cancellation**, toggle the setting on or off. 3. Click **Save**. ### Configure interstitial audio\[#interstitial-audio] During voice calls, there are moments when your AI Agent needs extra time to process a request — for example, when retrieving information from external systems. Without any audio feedback, callers may assume the call has dropped or the AI Agent has stalled, leading them to speak again or hang up. Interstitial audio plays a subtle audio cue during these processing moments, signaling to the caller that the AI Agent is still actively working. This reduces caller barge-in events, improves call quality, and creates a smoother overall experience. #### Choose a sound You can select from four purpose-built sounds, each designed and optimized for phone audio clarity: | Sound | Description | | ---------------- | --------------------------------------------------------------- | | **Calm Contour** | Reflective and steady, ideal for communicating deep thinking. | | **Bright Motif** | Minimal melodic ping that feels crisp, friendly, and efficient. | | **Pure Cascade** | Tonal sequence with a modern, focused feel. | | **Slow Gallop** | Light, syncopated pulse that feels calm and unobtrusive. | You can preview each sound in the dashboard before saving your selection. #### Set the delay threshold The delay threshold controls how long the AI Agent waits before interstitial audio begins playing. The timer starts after the caller stops speaking or after the AI Agent's acknowledgement message is spoken — whichever happens last. You can choose a delay between **2.0 and 4.0 seconds**. A shorter delay provides audio feedback sooner, while a longer delay avoids playing audio for requests that the AI Agent can resolve quickly. **To configure interstitial audio:** 1. On the Ada dashboard, go to **Config > CHANNELS > Voice**, then go to the **Interstitial Audio** tab. 2. Toggle **Interstitial Audio** on to enable the feature. 3. Under **Delay threshold**, select how long the AI Agent should wait before playing the sound. The default is **3.0 seconds**. 4. Under **Sound**, select one of the four available sounds. Click the **Play** button on any sound card to preview it. 5. Click **Save**. ### Improve recognition of special terms with Voice detection Brand-specific product names, internal acronyms, and industry jargon are easy for a speech-to-text engine to mishear. Flag those terms for Voice detection in [Glossary](/docs/automation/glossary), and the engine gives them extra weight whenever it transcribes a call. Glossary holds these terms alongside the preferred wording and definitions the AI Agent uses in every channel. The **Vocabulary** tab under **Config > CHANNELS > Voice** no longer lists terms. It links to Glossary. #### What makes a good term for Voice detection? | ✅ Do include | 🚫 Avoid | | ------------------------------------------------------------------ | -------------------------------------------------------------------- | | Brand names, product SKUs, internal feature names | Everyday words the engine already understands (“account”, “support”) | | Industry-specific jargon and acronyms | Full sentences or long paragraphs | | Short phrases, ideally under 20 characters (e.g., **“Flex-Plus”**) | Terms with punctuation the end user won’t say aloud | > **Note** > > Voice detection applies per language: on a call, the speech-to-text engine is sent the User terms you added for that call's language. Voice detection is not available for Albanian or Indonesian. > **Note** > > Up to 30 terms per AI Agent can have Voice detection enabled. Each speech-to-text engine also applies its own limits, and drops terms that are unusually long or that exceed the number it accepts. Keep terms short, and prioritize your most important brand and product names. The **Voice detection enabled** toggle appears only when your AI Agent has the Voice channel configured. **To flag a term for Voice detection:** 1. On the Ada dashboard, go to **Config > AI Agent > Glossary**. 2. Click an existing term, or click **New term** to create one. 3. Enable the **Voice detection enabled** toggle. 4. Click **Create** for a new term, or **Save** for an existing one. See [Glossary](/docs/automation/glossary) for the full term model. It covers translation modes, definitions, availability rules, and pronunciation on Voice. It also covers how to edit or delete a term, and how to find the conversations that used one. ### Test your Voice AI Agent\[#test-voice] Once Voice is enabled and a phone number is configured, you can test your AI Agent over voice from the dashboard, without deploying it or calling from an external phone. **To test your Voice AI Agent:** 1. Click **Test AI Agent**, then select **Voice**. 2. Choose how to connect: * **Web call**: the call runs in your browser, using your microphone and speakers. Useful for a quick check without a phone. * **Phone**: your AI Agent calls a number you enter, in international format, for example `+16135550183`. Use this to hear the call over a real phone line, as your end users will. 3. Select the language for the test. The list contains the languages you enabled for Voice. 4. Click **Call**. Test calls are recorded as test conversations and are excluded from your reporting. For details on the in-call controls and viewing the transcript, see [Interactive Testing](/docs/optimization/testing/interactive-testing#test-over-voice). ## Related features Explore additional Voice capabilities and related configuration options. * [Contact center integration](/docs/channels/voice/contact-center-integration): Integrate your Voice AI Agent with your CCaaS provider. * [Greetings and Handoffs](/docs/channels/voice/voice-configuration/greetings-and-handoffs): Configure how your Voice AI Agent greets callers and transfers them to agents. * [Voice tools](/docs/channels/voice/voice-configuration/voice-tools): Enable additional tools for your Voice AI Agent. * [Voice call capture options](/docs/channels/voice/voice-configuration/voice-call-capture-options): Configure how your AI Agent captures input by speech, keypad, or SMS. * [Languages](/docs/setup/languages): Configure multilingual support for your AI Agent. * [Interactive Testing](/docs/optimization/testing/interactive-testing): Test your AI Agent over voice or chat from the dashboard. * [Glossary](/docs/automation/glossary): Manage the terms your AI Agent recognizes and says, including Voice detection and Voice pronunciation. --- Have any questions? Contact your Ada team, or email us at [](mailto:help@ada.cx?subject=Help%20Docs%20inquiry).