Typecast
AISynthesize expressive speech, clone a voice from a sample, and get word-level timings with the Typecast API. Agents produce narration and voice prompts inside the workflow that needs them.
What This Integration Enables
Typecast is the connector for teams that need one voice to sound like a particular person, over and over, across content that keeps changing. That is a different proposition from picking a stock narrator. Clone Voice builds a custom voice from a recorded sample and then treats it like any other voice in the catalog, which means a presenter records once and their voice narrates a hundred modules they never sat down for.
The second thing Typecast gives an agent is timing. Generate Speech With Timestamps returns the start and end of every word or character alongside the audio, which is what a subtitle file, a karaoke highlight, or a lip sync rig actually needs. Compose Speech renders up to fifty ordered segments including silent pauses into one file, so multi-speaker dialogue and paced announcements are a single call rather than a stitching job. Synthesis is synchronous, so the audio is ready when the step completes and the next step in the flow can consume it directly. Because this connector can put words in a real person's mouth, the digital andon cord belongs on it by default, not as a setting somebody remembers to turn on.
Without FlowRunner
With FlowRunner
Use Case Scenarios
A course catalog narrated in one voice
A training team publishes modules continuously and wants them all to sound like the same instructor. The instructor records one clean sample, a person approves the clone, and Clone Voice registers it. From then on, each approved script is split into segments and pauses, rendered with Compose Speech, and attached to the module record. When legal changes a sentence six months later, the agent re-renders that module alone. The instructor's voice stays consistent across the catalog without the instructor being in the loop for every edit, and the approval on the clone itself is where their consent was captured.
Subtitles and lip sync built from the same render
A video needs narration and matching captions. The agent calls Generate Speech With Timestamps, gets the audio file plus the start and end time of every word, and builds an SRT file from the alignment array in the same step. The caption file and the audio file are generated together from one source of truth rather than reconciled afterward, then pushed to YouTube with the upload. The same alignment data drives character mouth timing when the audio feeds an animation pipeline.
Casting a voice for a campaign
Instead of browsing the full catalog, a producer describes what they want in plain language: a warm female narrator for a wellness explainer in English. The agent passes that brief to Recommend Voices, gets back a ranked shortlist, calls Get Voice Details to confirm which emotion presets each candidate actually supports, and renders the same fifteen second line in each. The producer picks from a handful of real samples rather than a list of names, and the chosen voice identifier gets written into the campaign record for every subsequent render.
Human-in-Loop Highlight
Voice cloning is where this connector earns a hard gate, and a generic approval step is not it. Clone Voice takes a recording of a real person and produces a voice that can say things that person never said. So the agent never calls it on its own initiative. When a flow reaches a clone step, it stops and sends the request to the person whose voice it is, by email so the reply is durable: "Rendering the Q4 module set in your voice. The sample is your March webinar intro, 41 seconds. The clone will be used for e-learning narration on this course catalog only, and it can be revoked. Reply approve to proceed, or reply hold." Nothing is cloned without that reply on file, and the reply becomes the consent record the flow checks on every later run.
The same gate has a second edge. Delete Custom Voice permanently removes a clone and frees the plan slot, and every flow still pointing at that voice identifier fails on its next synthesis call. Revocation is therefore also a human decision, routed to the voice owner and the content owner together, so the person taking their voice back and the team whose pipeline depends on it are looking at the same message.
Agent Capabilities
9 actionsSpeech Synthesis
3- Generate Speech Synthesize spoken audio from up to 2000 characters using a Typecast voice, stored in FlowRunner file storage and returned as a downloadable URL. Emotion preset and intensity shape the delivery while loudness, pitch, and tempo tune the result. Supplying a seed makes repeated runs with identical inputs reproducible, which matters when a flow re-renders the same line.
- Compose Speech Render an ordered list of spoken segments and silent pauses into a single file, which is how multi-speaker dialogue, guided meditations, and paced announcements get built in one call. Accepts between 1 and 50 segments with at least one spoken segment, up to 2000 combined characters and 60 seconds of combined pauses. Pause segments consume no credits.
- Generate Speech With Timestamps Synthesize speech and return alignment data giving the start and end time of every word or character, at the granularity you request. Used to drive subtitle files, karaoke highlighting, and lip sync from the same render that produced the audio.
Voice Management
2- Clone Voice Create a custom voice from a recorded sample so it can be used like any built in voice. The sample must be WAV or MP3, under 25 MB, and between 5 and 150 seconds. The returned identifier is bound to the model it was cloned for, and custom voice slots are limited by plan. Route this behind an approval from the person whose voice is being cloned.
- Delete Custom Voice Permanently remove a cloned voice and free the plan slot it occupied. Built in voices cannot be deleted. Any flow still referencing the deleted identifier will fail on its next synthesis call, so treat this as a coordinated revocation rather than cleanup.
Voice Catalog
3- Recommend Voices Suggest voices matching a free text brief describing style, mood, language, or use case, returned ranked by relevance. Used to shortlist candidates for a campaign without browsing the full catalog, requesting between 1 and 10 suggestions.
- List Voices Return the voice catalog available to the account, including stock voices and any clones, filterable by model support, gender, age group, use case, and voice type. Each entry lists the models it supports and the emotion presets available for each.
- Get Voice Details Retrieve the full profile of a single voice, including its classifications, the content types it suits, and which emotion presets each supported model offers. Used to confirm a voice supports an emotion such as Whisper before a synthesis call requests it.
Account
1- Get Subscription Report the current plan, monthly credit allowance, credits consumed, concurrent request limit, and how many custom voice slots the plan provides. Used to guard a batch render against running out of credits mid job or to alert when usage crosses a threshold.
Frequently Asked Questions
What can FlowRunner do with Typecast?
FlowRunner agents can run Generate Speech, Compose Speech, and Generate Speech With Timestamps in Typecast, plus 6 more actions.
Does connecting Typecast to FlowRunner require OAuth?
No. Typecast connects to FlowRunner with an API key, no OAuth flow required.
Can Typecast trigger a FlowRunner workflow automatically?
Typecast doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Typecast
$100 in credits. No card required. Connect in minutes.