Third-party large model integration for intelligent outbound calls

Integration solution

API definition

To integrate, follow the four interface conventions below.

Function Compute proxies these APIs, primarily handling signature authentication and forwarding.

On the intelligent outbound bot page, configure the bot by following the process described below.

• /proxy/beginSession: Begin session

  Description

  Starts a session.

  Protocol

  Server-Sent Events (SSE)

  Request parameters

  Parameter	Type	Required	Example	Description
  requestId	String	Yes	UUID	A unique ID for the request.
  tenantId	Long	Yes	378210	The tenant ID, which is your Alibaba Cloud UID.
  instanceId	String	Yes	n1	The instance ID.
  scriptId	String	Yes		The scene ID.
  sessionId	String	Yes	UUID	The session ID.
  streamId	String	Yes	UUID	The dialog turn ID.
  vendorParams	Map	Yes		Passthrough parameters.
  callingNumber	String	Yes		The calling number.
  calledNumber	String	Yes		The called number.
  Other fields	String	No		Customer-defined passthrough fields.

  Response parameters

  Parameter	Type	Required	Example	Description
  requestId	String	Yes	1	The unique ID of the original request.
  httpStatusCode	Int	Yes	200	The HTTP status code.
  code	String	Yes	OK	The business response code. A value of 'OK' indicates success.
  message	String	No		The error message that is returned if the request fails.
  data	Object	Yes
  streamEnd	Boolean	Yes		Indicates whether the stream has ended.
  updatedTime	Long	Yes		The time of the last update, in milliseconds.
  sessionId	String	No		The session ID.
  streamId	String	No		The dialog turn ID.
  answer	String	No		The text of the bot's response.
  controlParamsList	List<ControlParams>	No		A list of bot speech control parameters.
  type	String	Yes	HangUp	The control type. - HangUp: Disconnects the call. - Transfer: Transfers the call. - GatherDtmf: Collects DTMF input. - Voice: Plays a voice prompt.

• /proxy/dialogue

  Description

  Manages conversational turns in a dialogue.

  Protocol

  SSE

  Request parameters

  Parameter	Type	Required	Example	Description
  requestId	String	Yes	UUID	The request ID.
  tenantId	Long	Yes	231232121	The tenant ID. This is the Alibaba Cloud UID.
  instanceId	String	Yes	n1	The instance ID.
  scriptId	String	Yes		The script ID.
  sessionId	String	Yes	UUID	The session ID.
  streamId	String	Yes	UUID	The stream ID.
  utterance	String	Yes	Check phone bill	The user's utterance.
  extras	List<Map>			Additional data.
  dataType	String		DTMF	The data type. Valid values: DTMF or Silence.
  dtmf	String		1234	The DTMF digits.

  Response parameters

  Parameter	Type	Required	Example	Description
  requestId	String	Yes	1	The request ID.
  httpStatusCode	Int	Yes	200	The HTTP status code.
  code	String	Yes	OK	The response code.
  message	String	No		The error message.
  data	Object	Yes
  streamEnd	Boolean	Yes		Indicates whether the stream has ended.
  updatedTime	Long	Yes		The update timestamp.
  sessionId	String	No		The session ID.
  streamId	String	No		The stream ID.
  answer	String	No		The bot response.
  controlParamsList	List<ControlParams>			A list of control parameters.
  type	String	Yes	HangUp	The control type. - HangUp: Disconnects the call. - Transfer: Transfers the call. - GatherDtmf: Gathers DTMF input. - Voice: Plays a voice prompt.

• /proxy/abortDialogue

  Description

  Aborts the current dialogue turn. This operation returns any text that has already been played and stops all subsequent actions.

  Protocol

  HTTPS

  Request parameters

  Parameter	Type	Required	Example	Description
  requestId	String	Yes	UUID	The request ID.
  tenantId	Long	Yes	123232123	The tenant ID. This is the Alibaba Cloud UID.
  instanceId	String	Yes	n1	The instance ID.
  scriptId	String	No		The script ID.
  sessionId	String	Yes	UUID	The session ID.
  streamId	String	Yes	UUID	The stream ID.
  answer	String	No		The text that was played before the interruption.
  type	String	Yes	Canceled	Canceled: The turn is canceled. Interrupted: The bot is interrupted.

  Response parameters

  Parameter	Type	Required	Example	Description
  requestId	String	Yes	1	The request ID.
  httpStatusCode	Int	Yes	200	The HTTP status code.
  code	String	Yes	OK	The response code.
  message	String	No		The error message.

• /proxy/endSession

  Description

  Ends the session.

  Protocol

  HTTPS

  Request parameters

  Parameter	Type	Required	Example	Description
  requestId	String	Yes	UUID	The request ID.
  tenantId	Long	Yes	12321231	The tenant ID. This is the Alibaba Cloud UID.
  instanceId	String	Yes	n1	The instance ID.
  scriptId	String	Yes		The script ID.
  sessionId	String	Yes	UUID	The session ID.

  Response parameters

  Parameter	Type	Required	Example	Description
  requestId	String	Yes	1	The request ID.
  httpStatusCode	Int	Yes	200	The HTTP status code.
  code	String	Yes	OK	The response code.
  message	String	No		The error message.

• Common control parameters

  HangUpControlParams

  Parameter	Type	Required	Example	Description
  delay	Int	No	0	The hang-up delay in seconds. The default value is 0. Constraints: The value must be an integer from 0 to 10.
  reason	String	No		The reason for hanging up.

  TransferControlParams

  Parameter	Type	Required	Example	Description
  transferType	String	Yes	SkillGroup	The transfer type. - SkillGroup: Transfers to a skill group. - Agent: Transfers to an agent (currently unsupported). - External: Transfers to an external number (currently unsupported).
  transferee	String	Yes	skg-1234*****	The transfer destination.
  transferor	String	No		The transferor's phone number, which is required when transferring to an external number.

  DtmfControlParams

  Parameter	Type	Required	Example	Description
  terminator	String	Yes	#	The terminator for DTMF collection. Constraints: Only the # character is supported.
  maxDigits	Int	Yes		The maximum number of digits to collect. Constraints: The value must be an integer from 1 to 100.
  timeout	Int	Yes		The timeout, in seconds, between digit presses. Constraints: The value must be an integer from 1 to 10. A value of 5 is recommended.
  disableVoiceInput	boolean	No		Specifies whether to disable voice input during DTMF collection. Default value: true.

  VoiceControlParams

  Parameter	Type	Required	Example	Description
  interruptible	Boolean	No	false	Specifies whether the bot's speech can be interrupted. The default value is false.
  silenceDetectionTimeout	Int	No		The timeout, in seconds, for detecting user silence. The default value is 5. Constraints: The value must be an integer from 1 to 10. A value of 5 is recommended.
  synthesizerConfig	Object	No		Parameters for controlling the bot's synthesized voice. The following providers are supported: Alibaba Cloud small models Alibaba Cloud CosyVoice large models Doubao large models
  voice	String	Yes		The voice for speech synthesis. For a list of available voices, refer to the documentation of the corresponding provider.
  volume	Double	No		The speech volume. The default value is 50. For the supported value range, refer to the documentation of the corresponding provider.
  pitchRate	Double	No		The pitch rate. The default value is 0. For the supported value range, refer to the documentation of the corresponding provider.
  speechRate	Double	No		The speech rate. The default value is 0. For the supported value range, refer to the documentation of the corresponding provider.

Note

1. Create a Function Compute instance in the China (Shanghai) region. This lets you use a private IP address to reduce network overhead. You can disable the public IP address for enhanced security.

2. Each type of ControlParams has specific requirements:

   1. Voice: This parameter must be sent in the initial packet. It is ignored in subsequent packets.

   2. GatherDtmf: This parameter must be sent in the initial packet. It is ignored in subsequent packets.

   3. Transfer: This parameter can be sent at any time during the conversation.

   4. HangUp: This parameter can be sent at any time during the conversation.

3. Although the beginSession interface is designed for streaming, it is currently implemented as a synchronous blocking call. Therefore, the API response time must be under 200 ms.

4. To ensure a low-latency conversation, the large language model's time to first byte must be under 600 ms.

5. Consider the following for text-to-speech (TTS) playback of the large language model's responses:

   1. Format numbers in the text output to ensure they are read correctly by the text-to-speech engine.

   2. Some text-to-speech engines may struggle with the intonation of words with multiple pronunciations.

Function Compute demo

chatbot-proxy.zip

Configuration procedure

Prerequisites

1. Enable Function Compute.

2. Enable Cloud Outbound Bot.

Function Compute

1. Go to the Function Compute console: https://fcnext.console.aliyun.com/cn-shanghai/functions. In the top navigation bar, select China (Shanghai) as the region.

   

2. Create a Web function. This example uses a demo written in Java. The following figure shows the configuration:

3. Configure the function on its details page.

   1. In the trigger configuration, set the authentication method to signature method and enable the Disable public network access URL setting.

   2. Configure the Reserved Instance and horizontal scaling policies based on your business requirements. Set the number of Reserved Instances to at least 1 to prevent the first invocation from timing out.

4. Obtain the Web function's configuration.

   1. Function name

   2. Function region

   3. Trigger address. To minimize invocation latency, use the internal access address (HTTP).

Configure Cloud Outbound Bot

1. In large model scenario management, create a scenario and enter the function service parameters.

2. After creating the scenario, open it and configure the voice & VUI settings.

3. After completing the configuration, click voice debugging to run a test.

4. Return to the large model scenario list and click publish. Once published, you can use the scenario when creating a new outbound call task.