Voice Virtual Agent

Voice Virtual Agent

1. Configuration

1.1 Say (TTS) - Set "Voice"

1. In the editing interface, find the "Language/Region" option and click the drop-down menu on the right to select the desired language and region.

2. Find the "Series" option and click the drop-down menu on the right to select the appropriate "Series". Each series has different performance in different voice, price and delay.

3. Find the "Tone" option and click the drop-down menu on the right to select the desired tone. If you want to adjust the timbre, tone, volume, etc., please adjust the parameters under "Advanced Options.

4. If you need to set the background sound to simulate the real physical environment, select from the drop-down menu on the right side of "Background Sound.

5. Click the green play button to the right of "audition content" to audition.

6. After the audition is satisfied, click the "OK" button in the lower right corner to save the settings; If not satisfied, click the "Cancel" button to give up the settings.

7. After adjusting the "Voice" parameter, you can use it when creating a voice AI Agent.

 

 

1.2 Listen (ASR)

 

1.3 How to Work (LLM + Workflow) - Basic information of AI Agent

1. Enter the name of the AI Agent in the Name field.

2. Describe the working background and overall goal of the AI Agent in the "session background", or specify it directly in the subsequent intelligent dialogue.

3. Click on the "Language/Region" drop-down menu to select the language of the appropriate target audience.

4. Click the "ASR" drop-down menu to select the appropriate ASR option, each option is different in recognition rate and speed delay.

5. Click the "Voice" drop-down menu to select the sound type for debugging.

6. Turn on or off the "AI engine" according to whether AI capability is required. If it is turned off, AI-related nodes, such as "intelligent dialogue", cannot be used in subsequent processes.

7. Click the "Model Version" drop-down menu to select the appropriate model version. Currently, 4o and 4o-mini models of Open AI are supported. 4o-mini responds faster.

8. You can enter a description of the AI Agent in the description field.

9. Check that the settings are correct and then save.

 

 

 

2. Process Design (Workflow)

2.1 Process Overview

 

2.2 Node introduction

AI Agent Conversation

• Name: You can set a descriptive name for the node.

• Prompt words: you can enter requirements for LLM, and you can tell your employees what to do as you teach them.

• Optimization: You can provide optimization suggestion copy for your prompt words, but it is only for reference. You need to carefully debug the prompt words to meet your requirements.

• Fixed First Sentence: You can set a fixed statement at the beginning of the dialogue for this node. Fixed sentences help to unify the dialogue style and guide the direction of the dialogue.

Forced playback: The switch is turned on to ensure that the fixed first sentence is played. Otherwise, when the current dialogue has met the branching conditions, it will skip the current dialogue node and enter the next node. The fixed sentence of forced playback suggests to ensure that the dialogue is harmonious and coherent in this case and avoid abruptness.

• Branch name and description: identifies the full text of the current conversation. If the description is met, the current multi-round conversation will be withdrawn and the next node corresponding to the branch will be carried out. Please ensure that the conditions of the branch description are clear and independent.

• Path Tag: The switch opens the exit branch that can mark the conversation, and uses the branch name and branch description as "session tags". You can view these session tags in the session record. These session labels are easier to manage and analyze through reports

 

 

 

AI Agent Conversation - Advanced Settings

• User Interruption: Enable this switch to allow users to interrupt the conversation at any time.

• Response Timeout: Set the amount of time (in seconds) to wait for the user’s response.

• Maximum Turns: Set the maximum number of conversation turns. If the user continuously times out or the conversation exceeds the set number of turns, the system will exit the conversation early.

• Decision Description: Helps the agent determine the status of specific scenarios. For example, if the owed amount is 2000 and the customer pays 1000, it is considered a partial payment; if 2000 is paid, it is considered full payment.

• Global Jump: Enable this switch to allow any other “Intelligent Conversation” node to jump to this node. Useful for handling general questions outside the standard SOP (e.g., “Who are you?”).

• (Global) Branch Name and Description: Matches the description at any point in the conversation, providing flexibility to handle “other” topics.

When leaving the current conversation:

• Jump to another node: After ending a global node, directly jump to a specified node.

• Return to the previous node: After ending a global node, return to the previous node to continue the prior conversation.

 

Playback

1. In Name, enter a name for the play operation.
2. Select The Voice file acquisition method, select "text to voice" and enter the content after listening, or select "upload new voice" to upload the file.
3. In Wait Response Time, set the maximum time (seconds) to wait for the user to speak before executing the following process.

Waiting for response

• Name: Used to give an identifying name to the wait-for-response operation, here "wait-for-response".
• Response content storage variable: Specify a variable to store the content of the user's response (speaking).
• Wait time for response: Set the maximum number of seconds that the system waits for a response from the user.

 

 

Conditional Branch

Set multiple matching rules, and identify different conditions through "fuzzy matching", "exact matching" and "regular matching" of "variable" values to execute different branches.

 

 

Execution Period

It is used to perform related operations according to different days of the week and specific time periods in a specific time zone.

Proportional Branch

Distribute traffic to different branches according to A set proportion, which is often used in A/B test or traffic distribution scenarios.

 

Hang up

AI Agent will hang up after playing the voice.

 

Jump

After identifying the customer's intent, you can directly skip some unnecessary branch processes in the middle, which can make the interaction process smoother and more efficient, thereby improving the user experience.

SMS

The SMS service needs to be activated by contacting the account manager first.

Invoke API

Fill in the corresponding values according to your HTTP request parameters. You can obtain the data after the HTTP request in the response, and JSONPath syntax is supported.

The response body supports JSONPath syntax: JSONPath is a query language used to extract information from JSON data structures, similar to how XPath is used for XML. It allows you to use specific syntax (such as $ representing the root element, . for object members, and [] for array indices) to precisely locate and retrieve data.

RFC 9535: JSONPath: Query Expressions for JSONJSONPath Online Evaluator

Add Or Remove Tags

Variable Settings

Built-in Variables

3. Experience Testing

After the AI Agent process design is completed, click "Test" to start the voice call Test.

 

 

 

Voice Test

After entering the test page, you can see the delay of each sentence given by AI Agent in each link through the three line charts "TTFB(Time To First Byte)" on the right.

 

 

Text Test

You can also directly perform a text test to verify whether the process design meets expectations. The AI Agent's reply content is marked with the node from which it was sent to check whether the AI Agent's reply content and intent identification are reasonable.

 

 

Debug Mode

Click the debug icon to view the node flow process and decision results to help you better understand the execution process of the AI Agent.

 

4. Campaigns

Campaign Configuration

Operation path: Campaigns - Voice - AI Agent Outbound

Campaign Name

• Role: Used to give a unique identifier to the newly created campaign, which is convenient for users to identify and manage different campaigns.

AI Agent

• Role: Select the AI Agent to perform the campaign. Different AI agents may have different functions, language capabilities, etc.

Call Routing

• Role: Determines the path or mode of the call. Different routes may correspond to different call policies or targets.
• Example: Selecting "Default Route" means that the call will proceed in the default way preset by the system.

Keep Alive

• Function: Determine whether the campaign is a permanent campaign through the switch. Keep alive run continuously.
• Example: If enabled, the campaign will continue to run according to the set conditions.

Start mode

• Function: Determine the start mode of the campaign, with three options: manual, timing and immediate start.
  • Manual: The user needs to manually trigger the campaign to start.
  • Timing: You can set a specific time for the campaign to start automatically.
  • Start immediately: The campaign starts as soon as the setup is complete.
• Example: Manual is currently selected, meaning that user action is required to start this campaign.

Time zone

• Role: Set the time zone according to which the campaign is executed to ensure that the campaign is executed at the correct time.
• Example: Select "(UTC +08:00) Beijing, Chongqing, Hong Kong Special Administrative Region, Urumqi" to explain that the schedule of the campaign is based on the time of the eight eastern regions.

Dial cycle

• Role: Select which days of the week the campaign can be executed, providing flexible execution time settings.
• Example: Monday to Sunday is checked, indicating that this campaign may be performed every day.

Dial time slot

• Function: At an interval of 15 minutes, set in detail the specific time period during which the campaign can be performed every day.
• Example: By checking the green box, you can select a specific time period such as 9:00 - 9:15 AM to perform the campaign.

 

Import Number

Download the standard import template and fill in your number information. Red columns are required. Other columns correspond to the AI Agent variable list selected for the campaign. Fill in the actual values of these columns to help the AI Agent execute correctly.

Campaign Monitoring

RealTime Monitoring

1. Campaign monitoring area

• Role: Provides the overall running status of the campaign, allowing users to quickly understand whether the campaign is running.
• Content: includes the campaign status (such as running), Task ID, and task name.

2. Customer list area

• Role: Shows key customer-related metrics to help users understand customer-related business situations.
• Content: Including the number of customers, connection rate, call duration, call success rate and call connection rate.

3. Outcall Monitoring Chart

• Function: Show the trend of outbound call volume, connection rate and success rate over time in the form of charts to help users analyze the performance of outbound calls in different time periods.
• Content: Time range, outbound volume (green line), connection rate (orange line) and success rate (pink line) change curve.

4. Call duration distribution

• Function: Displays the distribution of call duration and helps users understand the concentration range of call duration so as to optimize call strategies.
• Content: the proportion and number of different call duration intervals (such as <1S, 1-10s, etc.), and the average call duration.

5. Call Results

• Function: The pie chart displays the proportion of different call results to help users intuitively understand the distribution of call results.
• Content: including the proportion of different results such as connection, customer not answering-ringing, customer not answering-not ringing, other-call failure, etc.

6. Call Tag

• Function: The Histogram displays the number of different call tags to help users understand the distribution of call tags for subsequent analysis and processing.
• Content: Including the number of different tags such as not connected, connected, not identified, and others.

This interface helps users to fully monitor and analyze customer calls during campaign execution through a variety of charts and data presentation methods, thereby optimizing business strategies.

 

Detail records

TOP OPERATING AREA

• Return: Used to return to the previous page.
• Campaign Name: displays the name of the current canpaign to help you identify the canpaign.
• Campaign Status: Displays the current status of the campaign (if paused) and has a refresh button to update the status.
• Start Campaign, Append Number, Campaign Configuration: These buttons are used to start, append number, and configure campaigns.

Campaign monitoring area

• Campaign Monitoring: This is a tab that may be used to switch to other campaign monitoring related views.

Customer List Area

• Customer List: This is a tab, currently selected, that displays customer-related information.
• Search bar: contains multiple search criteria, such as Enter name, enter number, select outbound call results, select session label, etc., used to filter the customer list.
• Search, Reset: Used to perform search operations and reset search criteria.

Customer Information Form

• Customer name, contact information, completion status, outbound call result, session label: These columns show the customer's basic information, campaign completion status, outbound call result, and session label.
• Summary, Cumulative Call Duration, Number of Calls, Number of Calls: These columns show statistics related to customer calls.

Customer Details Area

• After selecting a customer in the customer list, the detailed call history and content of the customer will be displayed below, including the conversation process and related questions between the customer and the customer.

This interface is used to manage customer lists, help users quickly find the customer information they need through search and filtering functions, and help users understand customer needs and problems through detailed call records.

 

Communication record

Conversation record

• Timestamp and message content: Each message has a sending time and specific content to show the sequence and content of the conversation.
• User avatar and identity: different users (may be customers and customer service) have different avatars and identities, which is convenient to distinguish.

Interaction Status

• User Interruption: A specific indicator displays the situation of the user interrupting the conversation to help the customer service agent understand the interaction details in the conversation.
• Timeout Prompt: There is a prompt to show the timeout of the user's operation, which helps to analyze the problem in the dialogue.

System Prompt

• Conversation round limit: There is a prompt to show that the conversation exceeds the maximum number of rounds, which helps the customer service to know whether the conversation meets the rules.
• Summary and intent labels: There are summary and intent labels at the bottom of the conversation to facilitate the customer service to quickly understand the key information of the conversation and the customer's intent.

This interface is mainly used for customer service to view the communication records with customers, through detailed time stamps, message content, interaction status and system prompts, to help customer service fully understand the dialogue situation, and make corresponding processing.