Skip to main content
Voice endpoint icon, a yellow letter V between sound waves
The Cognigy Voice endpoint enables you to connect your AI Agent to your contact center or other phone numbers using Cognigy Voice. This Endpoint allows your customers to speak with your AI Agents instead of using a chat.

Prerequisites

Generic endpoint settings

Learn about the generic Endpoint settings available for this Endpoint on the following pages. In the Endpoint, Copilot and Real-time translation are on the Behavior tab. Data protection and analytics and Session expiration are on the Compliance tab. NLU connector and Transformer functions are on the Advanced tab.

Endpoint-specific settings

The settings are grouped by the tabs of the Endpoint. Each accordion covers one tab.
EndpointChoose what answers the call, which speech providers it uses, and which phone numbers reach the Endpoint.TargetIf the Snapshot contains no agents, you can’t save an agent as the target. Select a Snapshot that contains an agent, or select No snapshot.SpeechThe Speech card defines the text-to-speech (TTS) and speech-to-text (STT) providers for calls on this Endpoint. Select from the speech providers in your project. To add a speech provider, see Speech and telephony.
The providers on this card are the defaults for calls on this Endpoint. If an agent defines its own speech settings, those settings take priority.The language and voice settings appear after you select a provider.If you select another TTS provider, Cognigy clears the language and the voice. If you change the TTS language, Cognigy clears the voice and the fallback voice. If you select another STT provider, Cognigy clears the STT language.
To use a fallback provider if the primary provider is unavailable, turn on Add fallback. The fallback uses the language of the primary settings. Only the provider and the voice change when the Endpoint switches to the fallback.If you change the target type or select a different agent, Cognigy clears the fallback.Number configurationIn the Number configuration card, assign the phone numbers of your project to this Endpoint.
  1. Click Add number.
  2. Select one or more phone numbers.
To remove a number, click the icon next to it. If a number is assigned to another Endpoint, click Reassign to assign it to this Endpoint. To add or manage numbers, click the links in the card. For more information, see Phone numbers.
The Voice endpoint includes the Click To Call embedding HTML code and the settings on the Click To Call tab for configuring Click To Call. Click To Call lets you embed a multimodal widget into your website, customer portal or mobile app so customers can be connected through one click to your voice AI Agent.AppearanceTranscription customizationPrivacy noticeDemo pageTo test the widget, click Open demo in new tab. The link works after you activate and save the Endpoint.Embedding HTMLTo embed the widget in your website, click Show embedding HTML snippet and copy the code. For more information, see Embed the Click To Call widget in your website.
Generic voice settingsThese settings are valid for every session of this Endpoint.
As with all Speech Provider settings, there are certain differences in the use of this feature. Refer to your Speech Provider’s documentation first to understand which values to add. An example would be the difference between Microsoft Azure and Google. For Microsoft Azure, users need to specify the % by which they would like to increase or decrease from the default settings. For instance, an example value could be Output speed -20%, indicating a 20% reduction from the default output speed of 100%, resulting in a new output speed of 80%. To achieve the same effect with Google, users need to add the final output speed in % to the Prosody settings. For example, Output speed 80% directly displays the new output speed, as with Google; the final output speed is directly added to the settings.
Call eventsCall events let you run an action in a Flow when a call event occurs. Select a call event from the Voice Events list. This event triggers the action.If you have configured the same call event in both the Endpoint and the Lookup Node, the Endpoint settings overwrite the Node settings.Call event settings
As with all other Endpoint settings, you can’t test the Call Events settings within the Interaction Panel.
Call failoverCall failover handles runtime and speech provider authentication errors that occur on the Cognigy side during the execution of a Flow.
For 2026.14.1 or earlier versions:
  • If you use a SaaS Cognigy installation, contact the support team to activate this feature.
  • If you use an on-premises Cognigy installation, activate this feature by adding FEATURE_ENABLE_ENDPOINT_CALL_FAILOVER in values.yaml.
For Cognigy Voice 2025.19 and later, the Anchor Media parameter has been renamed to Media Path and includes more options. The previously available options correspond to:
  • Full Media — the toggled on Anchor Media parameter in earlier versions.
  • Partial Media — the toggled off Anchor Media parameter in earlier versions.
The Media Path parameter is hidden behind the feature flag FEATURE_DISABLE_VG_MEDIA_PATH, with a default value of false. It is available only in environments where the feature flag is set to false.

Changing the media path during an active call

Besides setting the media path in the Flow, you can also change the media path during an active transferred call using the Cognigy Voice API.This approach is useful for PCI compliance. For example, when a caller shares sensitive payment information, Cognigy Voice, as an intermediate system, must be removed from the media stream. Audio is then routed directly between the caller and the contact center, bypassing Cognigy Voice, reducing PCI data exposure and helping protect sensitive data.To remove Cognigy Voice from the media path, send the PUT /Accounts//Calls/ API request with { "media_path": "noMedia" } from the contact center to Cognigy Voice after the call has been transferred. With this setting, Cognigy Voice stops processing the call’s audio and DTMF entirely, including for call transcription. After Cognigy Voice confirms the API request with a 200 or 202 response, you can’t revert to processing media for the transferred call. A 400, 404, or 422 response means the request failed, which keeps the call on its previous media path, and Voice continues processing audio and DTMF.Capture the call SID before the transfer takes place. During a transfer, a new call SID is created for the new call leg, but the original call SID remains unchanged. Pass the original call SID along during the transfer, for example, in SIP headers. This way, the contact center can reference the original call when triggering the API request at the correct moment.
Session expirationSession expiration sets the number of minutes that a session stays alive when it is idle. Enter a whole number, 0 or higher. Enter 0 so that the session never expires. An empty field is treated as 0.
Call recordingCall recording records calls on this Endpoint to a bucket that you configure in AI & Voice Providers. The setting is off by default and appears only if call recording is enabled for your installation.If your project has no recording buckets, add one in Manage > AI & Voice Providers > Telephony > Recordings, and then select it here. To open the page, click Open AI & Voice Providers.Transformer functionsTransformer functions define when the Endpoint runs a transformer function and what the transformer does.MockingIn the Developer section, Mocking replaces Voice calls with deterministic mocks. Use it only for local development. Don’t turn it on in production.

How to set up

Configure in Cognigy

  1. In the left-side menu of your Project, click Deploy > Endpoints.
  2. On the Endpoints page, click + New Endpoint.
  3. In the New Endpoint section, select the Voice endpoint type and enter a name.
  4. Select the agent or Flow that answers the call, and then save the Endpoint.
  5. On the Setup tab, configure the target, the speech providers, and the phone numbers. For more information, see Endpoint-specific settings.

Configure in your carrier or SIP trunk

To route calls from an external carrier or SIP trunk, send the traffic to the Endpoint URL on the Setup tab.

Additional information

The SIP headers, including any custom headers, are available within the Input object. You can find them in input.data or input.data.payload.sip.headers.
Voice identifies information about the caller and adds it to the Input object as input.data.numberMetaData.
The data in input.data.numberMetaData is available as Tokens inside Cognigy Text fields.
You can use any handover provider within the Voice integration. Handover enables your voice agent to route calls to the contact center, ensuring efficient communication and swift access to human assistance.In contrast to chatting, communication between actors during a voice call occurs using speech recognition technologies (text-to-speech and speech-to-text). Using speech-to-text technology, the voice agent sends the conversation script from the end user to the human agent. Once the script is received, the human agent can respond using text. After that, the voice agent converts the human agent’s message to speech using text-to-speech technology. Finally, the converted message is sent to the end user.ConfigurationTo enable handover, add a Handover to Human Agent Node in your voice Flow. This Node transfers the conversation from the voice agent to the human agent.In this example, an end user seeks billing assistance and interacts with agents in a customer service system via text-to-speech and speech-to-text technologies.End User: Calls the customer service line. Speaks to the voice agent. Example: Hello, I need assistance with my billing inquiry.
Voice Agent: Transcribes user’s speech into text using speech-to-text technology. Example: Hello, I need assistance with my billing inquiry.
Voice Agent: Determines the need for human assistance. Initiates handover to the human agent via handover provider.
Human Agent: Receives a conversation script containing user’s query.
Human Agent: Types response on interface. Example: Hello, thank you for reaching out. Let me check your billing details.
Voice Agent: Receives human agent’s typed response. Converts the text message into speech using text-to-speech technology.
End User: Hears human agent’s response spoken by voice agent. Example: Hello, thank you for reaching out. Let me check your billing details.
End User / Human Agent: The conversation between the end user and the human agent continues until the issue is resolved or the interaction is concluded.
Multilingual supportDuring a handover, you can use translation into other languages. Use the translation feature to overcome language barriers between end users and support staff. This feature is particularly useful in situations where there are no support staff proficient in the end user’s language, and urgent issue resolution is necessary.For example, if a user communicates in German, the voice agent responds accordingly. However, after transferring the call to the contact center, the dialogue is translated into English. The human agent responds in English, and the response is translated into German for the end user via text-to-speech technology. This method ensures seamless communication despite language differences.Note that depending on the language and provider selected, translations may experience delays while they reach the contact center. If the speech-to-text fails to recognize speech for any reason, the message won’t appear on the contact center’s side. Plan for such scenarios within your Flow and configure the voice agent’s behavior accordingly. This approach will ensure that the end user is aware that their message may not be delivered. You can achieve this by adjusting timeout settings and error handling policies.To set up a multilingual Flow for the handover use case, follow these steps:
  1. In your voice Flow, navigate to the Set Session Config Node.
  2. In the Node editor, go to the Recognizer STT section and activate the Recognize Language toggle.
  3. From the Alternative Language list, select the additional languages, for example, German and Italian.
  4. Below the Set Session Config Node, add a Lookup Node.
  5. In the parent Lookup Node, select CognigyScript from the Type list and specify input.data.payload.speech.language_code in the Operator field.
  6. In the case Nodes of the Lookup Node, specify the language codes, for example, de-DE, it-IT.
  7. Below each case Node, add the second Set Session Config Nodes.
  8. Configure the TTS and STT settings for each of the Set Session Config Nodes. This means that one Node should have German language selected, while the other should have Italian. These Nodes will manage the voice agent’s communication in the end user’s language. If you use an AWS voice, you can also select a TTS Engine. For more information, see Parameter Details.
  9. Below each Set Session Config Node, add the Set Translation Nodes. These Nodes will perform translation of a text from the end user’s language to the human agent’s language.
  10. In the Set Translation Node, activate the Translation Enabled toggle.
  11. In the User Input Language field, specify the language of the user input, for example, de, it.
  12. In the Flow Language field, specify the language to translate to, for example, en.
  13. At the end of the Flow, place the Handover to Human Agent Node.
In this example, an end user seeks billing assistance and interacts with actors in a customer service system via text-to-speech and speech-to-text technologies. The end user and human agent speak different languages, but this doesn’t hinder their understanding, as the voice agent translates their speech into the required language for each party.End User: Calls the customer service line. Speaks to the voice agent. Example: Hallo, ich brauche Hilfe bei meiner Rechnungsanfrage.
Voice Agent: Determines the language and answers using the same language. Example: Danke für die Informationen. Ich werde Sie an einen menschlichen Agenten weiterleiten, der Ihnen weiterhelfen kann.
Voice Agent: Transcribes user’s speech into text using speech-to-text technology. Example: Hello, I need assistance with my billing inquiry.
Voice Agent: Determines the need for human assistance and initiates the handover to the human agent via the handover provider.
Human Agent: Receives a conversation script containing user’s query.
Human Agent: Types response on interface. Example: I've found the issue. It seems there was an error in the billing statement. I'll correct it for you.
Voice Agent: Receives human agent’s typed response. Converts the text message into speech using text-to-speech technology.
End User: Hears human agent’s response spoken by voice agent. Example: Ich habe das Problem gefunden. Es scheint, dass es einen Fehler in der Rechnungsstellung gab. Ich werde es für dich korrigieren.
End User/Human Agent: The conversation between the end user and the human agent continues until the issue is resolved or the interaction is concluded.

More information

Last modified on November 16, 2026