Caller says “check my delivery” or presses 2.
Inbound flows that listen, act and respond.
Let callers answer a focused question by voice or keypad, route on a repeatable decision, use live business data and hear the result. Every fallback is visible before you publish.
Designed for short, structured English interactions. For open-ended conversations, use AI Agent Studio.
Send approved inputs and return named outcomes.
“Your driver is due between 2:10 and 2:25.”
No match, timeout and provider errors each have a designed route.
One journey across speech, connectors, routing, history and reporting.
Make one clear decision, then finish the task.
Speech Input does not send a caller’s words to an LLM to invent a route. It matches a committed short response against your configured phrases, then follows the named branch on the flow.
Listen
Ask one focused question. Accept a short spoken answer, a configured keypad key, or either.
Decide
Match an intent, confirm selected sensitive choices and follow an explicit route.
Act
Route directly or ask Connector Studio to check a booking, status, case or customer.
Respond
Speak a dynamic result, continue the service or use the relevant safe fallback.
Current scope: short, structured English-language interactions. This clear boundary keeps routing repeatable and makes it easier to choose Agent Studio when the caller needs a genuine conversation.
Generate what changes. Recognise what matters.
Speak Text and Speech Input are independent Flow Builder nodes. Use either one on its own, or combine them around Connector Studio for a complete self-service loop.
Speak Text
Generate and play a message when the flow reaches the node. Insert route-safe call data, an earlier speech result or an approved Connector output, then choose the voice and speaking speed.
- Best for: personalised or live-data messages.
- Author controls: text, available voice, speed and optional fallback audio.
- Required routes: Continue, Timeout and Error.
Speech Input
Play one complete prompt while listening for a short response. Map natural phrases and optional keys to named intents, then expose the selected result to later steps.
- Best for: focused choices with predictable routes.
- Author controls: phrases, keypad keys, listening time, retries and per-choice confirmation.
- Required routes: intents plus explicit no-input, no-match or exhausted, Timeout and Error handling as applicable.
Give people more than one way to succeed.
A speech journey should feel easier than the keypad tree it replaces, without removing the clarity operators rely on.
-
01
Speech or keypad
Match optional keypad numbers to the same choices as spoken phrases.
-
02
Natural interruption
Let callers start speaking during the prompt, or ask them to wait when clarity matters.
-
03
Guided retries
Play a short reminder, repeat the full question, or use both across one to three attempts.
-
04
Confirmation when it matters
Ask for a simple Yes or No only before sensitive choices continue.
-
05
Clear fallback routes
Choose what happens when there is silence, no clear match, a timeout or a service problem.
“Check my booking” follows the configured booking-status intent.
Use route-safe inputs with a pinned connector action revision.
Expose only the values and named provider outcomes the flow expects.
Use the reachable output variables in a personalised caller response.
Turn “press one” into “your booking is on its way.”
Connector Studio gives a call flow a controlled way to use live business systems. The speech layer makes those results useful to the caller in the same interaction.
- Install a reviewed provider template or define a reusable custom connector.
- Keep credentials encrypted, write-only and separate from revisioned connector configuration.
- Pin a flow to the connector revision it was reviewed against.
- Map provider results to named business outcomes, with mandatory Timeout and Error routes.
- Expose reviewed outputs to later flow nodes instead of passing an unbounded response through the call.
Know what will happen before and after you publish.
Flow Builder is not just a drawing surface. Simulation, immutable publish history and Call Journey reporting make the authored graph operationally useful.
Author the complete route
Flow validation checks connections, speech settings, route-safe variables and required success or failure ports before live publication.
Simulate without side effects
Walk a draft or saved service with example caller choices and connector results. Simulation does not place a call, invoke a provider or run the live business action.
Publish as a deliberate revision
A successful semantic change creates an immutable full-state service version. Compare changes, load an older revision as a draft and publish a restore without erasing the intervening history.
Follow the real Call Journey
Live Statistics and cradle-to-grave reporting show speech generation, listening, selected intent, interruption timing and safe failure outcomes without exposing the recognised transcript in routine journey telemetry.
Choose the smallest tool that solves the call.
Use deterministic speech when the job is narrow, an AI agent when the conversation is open-ended, and a fixed IVR when a recording and keypad are enough.
| Capability | Speech-enabled Flow Builder | AI Agent Studio | Fixed recorded IVR |
|---|---|---|---|
| Best interaction | Short phrase or keypad choice with a known route | Open-ended, multi-turn conversation | Fixed menu and announcement |
| Routing model | Repeatable intent match to explicit ports | Authored agent logic using conversation context | Digit-to-destination mapping |
| Dynamic spoken data | Speak Text with call, speech and Connector variables | Agent response informed by knowledge and actions | Requires separately generated or recorded prompts |
| Failure design | Named no-input, no-match, exhausted, timeout and error paths | Authored escalation and action or transfer failure paths | Timeout and invalid-key routes |
| Change control | Flow simulation, validation, compare and immutable versions | Agent test, workspace analytics and controlled deployment | Core service configuration and history |
A single CallPlatform service can combine these patterns and hand off into normal destinations, queues, voicemail, linked services or people where the journey requires it.
Start where the answer exists but the caller still has to ask.
Structured inbound automation works best when the question is frequent, the data is available and the exception route is clear.
“Where is my driver?”
Recognise the booking request, retrieve the current status and speak the approved ETA or next step.
“I need billing.”
Replace a long menu with a focused spoken choice while retaining keypad options and explicit team routes.
“When is my visit?”
Use caller and captured context to retrieve a record, then read back only the route-approved result.
“Cancel my booking.”
Confirm the selected intent before invoking a controlled action, and route exceptions to a person.
Personalise the answer
Use Speak Text for reference numbers, current dates, names and provider results that cannot live in one recording.
Design the human route
Send low-confidence, no-match or provider-failure cases to the right queue, service, voicemail or destination.
Clear boundaries make better call flows.
What is Speak Text?
Speak Text is a Flow Builder node that generates and plays a message during a live call. It can include route-safe call data, an earlier Speech Input result and approved values returned by Connector Studio. Every published node has Continue, Timeout and Error routes.
What can Speech Input recognise?
The current release is designed for short English responses mapped to configured intents. Authors provide natural example phrases and may assign a keypad key to each intent. It is not intended for unrestricted dictation or open-ended conversation.
How do retries, confirmation and barge-in work?
A node can allow one to three attempts, use a short retry prompt, repeat the full prompt, or do both. Confirmation applies only to choices marked as sensitive. Prompt interruption is enabled by default so callers can begin speaking while the prompt plays, and can be disabled per node.
Is Speech Input the same as an AI voice agent?
No. Speech Input matches a short committed response to configured intent phrases and follows an explicit route; routing is not delegated to an LLM. Agent Studio is the better fit for an open-ended, multi-turn conversation.
Can a flow use live data from another system?
Yes. A Connector Studio action can receive approved inputs, return reviewed outputs and follow named provider outcomes. A later Speak Text node can use outputs that are guaranteed to exist on that route.
What happens when speech or a provider fails?
Failure is part of the graph. Speak Text uses explicit Timeout and Error routes and can have fallback audio. Speech Input provides routes for no input, no match or exhausted attempts where applicable, plus provider Timeout and Error. Those paths can continue, transfer, use a recording or reach another CallPlatform service.
We’ll turn it into a journey you can see and test.
Map the caller’s words, the live system involved, the safe outcomes and the moment a person should take over. Then see how Flow Builder, speech and Connector Studio work together.