This topic describes feature updates for Artificial Intelligence Cloud Call Service.
Communication intelligence engine feature updates
Feature name | Change type | Description | References |
Configuration for transferring to a human agent | New | Automatically transfers calls to an agent platform when the communication intelligence engine receives a "transfer to human agent" tag from the Large Language Model (LLM). | |
Query TTS voice list | New | Retrieves the list of Text-to-Speech (TTS) voices supported for LLM-powered outbound calls, including preset system voices and custom voices. | ListAvailableTts - Get the list of TTS voices supported for LLM-powered calls |
Query dialog content | New | Queries dialog content by CallId. This feature only supports querying dialog content for calls that ended within the last month. | GetCallDialogContent - Query the dialog content of calls that ended within the last month by CallId |
Dialog interruption filler word policy | New | When a call is interrupted, a condition triggers filler words such as "Uh-huh, go ahead" or "Hmm" to create a smooth transition. | - |
Configuration for audio mixing and background sound | New |
| |
Morse code playback configuration | New | When enabled, a Morse code declaration generated by AI is played at the end of the conversation. | |
Smart answer detection | New | When enabled, automatically detects voice assistants or mailboxes and returns a smart status code. Configure whether to end the call when a voice assistant or mailbox is detected, as needed. | |
Inbound call feature configuration | New | Adds inbound call capabilities. Configure an inbound number for an LLM application to use the communication intelligence engine for inbound scenarios. | |
Communication resource integration | New | Supports networking modes with various resource compositions to provide customers with comprehensive communication network solutions. | - |
Call experience | New | The call experience feature lets you quickly try out LLM-powered intelligent calling. There are two ways to experience it:
| |
In-call denoising | New | This feature is being rolled out gradually based on internal user system IDs and is not yet fully available. | - |
Call API operation supports specifying a voice | New | The call API operation supports specifying a voice. You can query the list of available voices by PID. | |
Personalized adaptation for SipX playback | New | When you use a SipX number to call a callee, the carrier plays an audio clip after the call connects. You can personalize the SipX playback. After requesting a number from Voice Service, you must register it offline. | - |
Pass variables to the LLM | New | The LlmSmartCall API operation adds the `BizParam` field to support passing variables to the LLM. | |
Custom queues | New | Supports configuring a separate message queue or address for each application. | |
AXN mode | New | Supports a mode where users can dial an extension number. | - |
Maximum call duration | New | Set the maximum duration for a call. | |
Push call recording receipts | New | Configure receipt messages for call recordings to get information such as the call start time, end time, duration, and recording file URL. |
|
Breakpoint calculation for interruptions | New | Calculates the breakpoint to determine what content has been spoken when an interruption occurs. | |
Barge-in model optimization | New | Optimizes the barge-in model to wait for the customer to speak first. | - |
Integration with Model Studio models and applications | New | Supports integration with Alibaba Cloud Model Studio models and applications. You can connect to model APIs and application APIs. | |
Voiceprint replication | New | Customers can upload recording files in the console to complete voiceprint replication as a self-service feature. | |
Add extension field to LlmSmartCall | New | The `LlmSmartCall` API operation's response parameters include the `OUTID` business extension field. This is a custom parameter that can be output during model interaction. | |
Hang up by tag | New | Hang up a call using the `<hang-up>` tag. The model outputs the tag, and the engine automatically executes the event after detection. No manual configuration is required. | - |
Do not interrupt on filler words | New | When using the communication intelligence engine, recognized filler words such as "uh-huh," "okay," and "go on..." do not trigger an interruption. | - |
Communication agent feature updates
Feature | Change type | Description | References |
Configuration for audio mixing and background sound | New |
| |
Voiceprint replication | New | Customers can upload recording files in the console to complete voiceprint replication as a self-service feature. | |
Opening statement type supports recording files | New | Supports playing a specified recording file as the opening statement after the call connects. | |
Smart answer detection | New | When enabled, automatically detects voice assistants or mailboxes and returns a smart status code. Configure whether to end the call when a voice assistant or mailbox is detected, as needed. | |
Export task details | New | Supports exporting the details of completed tasks from the console. | |
Task management and receipt push | New | Call task management APIs and push receipt messages. |
|