Collect Input Node (DTMF and Speech-to-Text)
Overview
The Collect Input node enables flow builders to gather user input during interactions and store the captured value in a flow variable for use in downstream logic. It provides a deterministic method of collecting information without relying on LLMs, intent detection, or NLU processing.
The node supports:
DTMF (keypad) input
Speech input using Speech-to-Text (STT)
Combined DTMF + Speech mode
The node is designed to capture exactly what the user enters or says and store it as a variable. It does not interpret meaning, infer intent, extract entities, or invoke AI reasoning.
Key Benefits
Deterministic Input Collection
The Collect Input node provides a reliable way to gather user-provided values such as:
Account numbers
Order IDs
Reference numbers
Postal codes
Callback numbers
Customer names
Verification information
Unlike AI-driven interactions, the node captures information exactly as provided and follows predefined validation rules.
Reduced Cost and Latency
For many voice journeys, transcription is all that is required.
The Collect Input node:
Uses DTMF and/or STT only
Does not invoke LLM reasoning
Reduces processing latency
Lowers operational cost
Provides predictable behavior
Reusable in Flow Logic
Captured values can be used throughout the flow:
Branch conditions
API Calls
Set Variable nodes
Validation steps
Routing decisions
Agent handoff context
Supported Input Modes
1. DTMF Mode
DTMF mode captures keypad input entered by callers.
Best For
Account numbers
PINs
Member IDs
Case references
Numeric order IDs
Callback phone numbers
Available Settings
Minimum Digits
Minimum number of digits required
Maximum Digits
Maximum number of digits allowed
Terminator Key
Key used to indicate input completion
Inter-Digit Timeout
Maximum time allowed between key presses
Retry Count
Number of retry attempts allowed
Timeout Message
Prompt played when timeout occurs
Invalid Input Message
Prompt played when input validation fails
Example
Prompt: Please enter your 4-digit account number.
Configuration:
Minimum Digits: 4
Maximum Digits: 4
Terminator Key: #
Retry Count: 2
Accepted Input: 1234
Rejected Input: 12
The node will play the configured retry prompt until retries are exhausted.
2. Speech Mode
Speech mode captures caller speech and converts it into text using the Speech-to-Text provider configured in the flow's voice settings.
Important
The node:
Uses STT only
Does not invoke an LLM
Does not interpret user intent
Stores the transcript exactly as recognized
Example Use Cases
Customer names
Short reference phrases
Spoken identifiers
Callback requests
Example
Prompt: Please say your full name.
Caller says: John Doe
Stored variable: John Doe
3. DTMF + Speech Mode
This mode allows users to provide input either through:
Keypad entry
Speech
Example Prompt
You can enter or say your account number.
Runtime Behavior
The first valid input method is accepted.
Example:
User starts entering DTMF digits
DTMF collection becomes active
Speech input will not interrupt the ongoing DTMF collection
If the user wishes to change input method:
Complete the DTMF collection using the terminator key
Retry the prompt
Provide speech input
DTMF Configuration
Minimum and Maximum Digits
These settings define valid input length.
Fixed Length Input
Example: 4-digit account number
Minimum Digits
4
Maximum Digits
4
Only 4 digits are accepted.
Variable Length Input
Example: Account number can contain different lengths.
Minimum Digits
4
Maximum Digits
12
Any value between 4 and 12 digits is considered valid.
Terminator Key
Specifies the key used to indicate completion of input.
Default: #
Other supported options may be selected from configuration.
Example: 123456#
The system treats the input as complete when the terminator key is pressed.
Inter-Digit Timeout
Defines how long the system waits between two key presses.
Example:
If a user enters: 1
and waits longer than the configured timeout before entering the next digit:
Input times out
Retry prompt is played
Retry Handling
The node supports configurable retry behavior.
Invalid Input
Occurs when:
Input does not meet minimum length
Input exceeds maximum length
Input fails configured collection rules
Example:
Expected: 4 digits
User enters: 12
The node:
Plays retry prompt
Waits for new input
Repeats until retry limit is reached
No Input
Occurs when:
User provides no response
Timeout period expires
The node:
Plays no-input prompt
Allows retry attempts
Follows the failure branch if retries are exhausted
Max Retries Exhausted Branch
Once retry limits are reached:
Input collection stops
Flow follows the Max Retries Exhausted path
This enables builders to:
End the flow
Transfer to an agent
Route elsewhere
Provide escalation messaging
Runtime Behavior
Successful Input
When valid input is captured:
Value is saved to the configured variable.
Flow proceeds through the success path.
Downstream nodes can consume the value.
Failed Input
When input validation fails:
Retry prompt plays.
Retry counters increase.
Failure path executes after maximum retries.
Voice Settings Dependency
Speech-based collection uses the Speech-to-Text provider configured at the flow level.
Navigate to:
Flow Settings → Voice Settings
The Collect Input node uses the configured:
Speech-to-Text provider
Text-to-Speech provider
Voice configuration
No separate STT model selection is required within the node itself.
Barge-In Behavior
Speech Input
If barge-in is enabled, callers can interrupt the system prompt using speech while the prompt is playing.
DTMF Input
Once a DTMF collection session has started:
Speech cannot interrupt the active DTMF collection.
Input modes cannot be switched mid-collection.
Users must complete or terminate the current DTMF entry before providing speech input.
AI Usage
The Collect Input node is intentionally designed as a deterministic collection mechanism.
What Uses AI?
Only Speech-to-Text transcription.
When a caller speaks:
Speech is transcribed using the configured STT provider.
The resulting transcript is stored.
What Does Not Use AI?
The node does not:
Detect intent
Extract entities
Classify sentiment
Apply reasoning
Invoke LLMs
Perform NLU processing
The collected value is stored exactly as entered or transcribed.
Transcript Variable Support
In the same release, Flow Designer introduces an official Transcript Variable.
The transcript variable:
Captures the complete conversation transcript
Follows the configured PII redaction policy
Redacts sensitive information when redaction is enabled
Can be sent to external APIs for processing or analytics
Example Use Cases
Conversation analysis
Knowledge extraction
Quality monitoring
External AI processing
Analytics workflows
PII Redaction Support
When PII redaction is enabled:
Personal information is automatically redacted in the transcript variable.
The transcript remains safe for downstream sharing and processing based on configured policies.
Best Practices
Use DTMF for structured numeric values.
Use Speech when spoken input is more natural.
Use DTMF + Speech only when flexibility is required.
Clearly communicate expected input format in prompts.
Configure retry messages that guide users to correct mistakes.
Validate captured values downstream where business validation is required.
Use meaningful variable names such as
accountNumber,orderReference, orcustomerName.Configure appropriate timeout and retry values to improve completion rates.
Summary
The Collect Input node provides a simple, reliable, and deterministic way to capture user input in voice journeys. It supports DTMF, speech, or both input methods, stores the captured value in a flow variable, and integrates seamlessly with downstream flow logic. By relying only on keypad input and Speech-to-Text transcription, it offers lower latency, lower cost, and predictable execution while giving builders a flexible foundation for common IVR and voice automation scenarios.
Last updated
Was this helpful?