FlowRunner
PricingContact
Theme
Start Free

PlayHT

AI

Give AI agents an ultra-realistic voice with PlayHT text-to-speech. Agents generate lifelike speech in 36 languages from hundreds of stock voices, clone a brand voice from an audio sample, and store finished audio in FlowRunner file storage with ready download URLs.

7 actions API key available
PlayHT website ↗ Platform Documentation ↗ Capability data verified 2026-07-25
Release notes are finalized in [Notion](/integrations/notion)
List Cloned Voices resolves the brand voice's manifest id
Generate Speech renders the narration and stores the finished audio in FlowRunner file storage
The agent confirms the job finished and that the audio duration is plausible for the script length
The product marketer listens through, checking product names, version numbers, and tone, before release
The approved audio embeds in the release post and its file URL logs to the content tracker
The launch channel in [Slack](/integrations/slack) gets the listen link

What This Integration Enables

PlayHT's distinctive move is treating a voice as reusable infrastructure: clone it once from a clean sample, and every workflow that narrates, announces, or answers can speak as the brand. The engine lineup is the other half of the story. PlayDialog drives conversational prosody for dialog and briefings, Play 3.0 Mini handles 36 languages and long inputs with reliable numbers, and the classic engines carry fine-grained emotion and guidance controls. The connector wires all of it into flows and stores every finished file in FlowRunner file storage with a ready download URL, so no render is ever a temporary link. A voice is also an identity, which is why the operations that create and destroy voices run [human-in-the-loop](/concepts/human-in-the-loop). - Narrate articles, briefings, and release notes in a consistent cloned brand voice - Pick engines per job: conversational, multilingual long-form, or guidance-controlled - Batch-render narration asynchronously and collect finished audio as jobs complete - Manage the voice catalog, stock and cloned, from within flows and with consent on record

Without FlowRunner

The brand voice is one person's calendar Narration waits for whoever records it to have a free hour and a quiet room
Long scripts fight the tooling Legacy text-to-speech chokes on length, numbers, and product names
Localized audio means agencies Each language is a procurement exercise, so most content ships in one

With FlowRunner

The brand voice is an asset Cloned once from a sample, applied by any workflow that needs it
Engines match the job Conversational prosody for dialog, a long-form multilingual engine for documents
Every render lands in storage Audio saves to FlowRunner file storage with a ready URL, not a temporary link

Use Case Scenarios

Every article gets a listen link

A post publishes on [WordPress](/integrations/wordpress) and the narration flow takes over: Generate Speech renders the article in the brand's cloned voice, the stored file uploads back to the site as a listen-to-this-post attachment, and the editor gets the review link first. Long pieces route through Stream Speech on the Play 3.0 Mini engine, which takes much longer inputs in one pass and handles the version numbers and product names that reliably trip lesser engines.

The morning briefing that speaks

The agent compiles the daily operations digest, then voices it through Stream Speech with the PlayDialog engine, whose conversational pacing keeps a two-minute briefing from sounding like a robot reading a spreadsheet. The audio lands in [Slack](/integrations/slack) before the day starts, and the team absorbs it during the commute instead of reading it at the desk. The same pattern serves spoken alerts: a threshold breach becomes a short voiced message in the on-call channel, immediate to consume with hands and eyes busy.

Back-catalog narration as a managed batch

Hundreds of lesson scripts sit in [Airtable](/integrations/airtable). The agent submits each through Generate Speech with Wait For Completion disabled, collecting job ids instead of blocking, then polls Get TTS Job and writes each finished file URL back to its row. Failures surface in a retry queue with reasons attached. The course team reviews samples from each voice before the batch publishes, and the whole catalog gains audio without anyone babysitting a render queue. When a script changes later, re-rendering one lesson is one job, not a project.

Human-in-Loop Highlight

Two operations here deserve a named approver. Clone Voice creates a working replica of a real person's voice from a single audio sample, and that is a consent decision before it is a technical one: whose voice, authorized by whom, for which uses, gets decided by a person and recorded, not inferred by a workflow. Delete Cloned Voice is the other edge: it permanently removes a clone, and any flow still referencing the deleted id fails on its next speech request. An agent tidying up voice slots can silently break every narration pipeline built on that voice. So deletions present which flows reference the id and wait for a human to accept the blast radius, and no clone is created without a documented owner and consent.

Agent processes routinely
Detects exception requiring judgment
Clear match Continues automatically
Ambiguous Routes to human via preferred channel
Human decides
Agent resumes with decision

Agent Capabilities

7 actions

Speech Generation

3
  • Generate Speech Converts text to speech through the asynchronous job API, polling until rendered, then downloading and storing the audio in FlowRunner file storage with a ready file URL. Works with stock and cloned voices; the PlayHT 2.0 and 1.0 engines take roughly 2,000 characters per request. Disable Wait For Completion for fire-and-forget batches.
  • Stream Speech Generates speech through the low-latency streaming endpoint, returning audio directly and storing it with a file URL. The path to the newer engines: PlayDialog for conversational prosody, Play 3.0 Mini for 36 languages and inputs up to roughly 20,000 characters, and PlayHT 2.0 Turbo.
  • Get TTS Job Retrieves an asynchronous job's status and settings, and once rendering completes, its duration, size, and download URL, optionally storing the audio. The tracking half of batch pipelines.

Voice Management

1
  • List Voices Retrieves the stock catalog of several hundred voices with manifest id, language, gender, accent, age, style, and a preview sample, filterable by search text, language, and gender.

Voice Cloning

3
  • Clone Voice Creates an instant clone from an audio sample fetched from a URL. Samples must run between 2 seconds and 1 hour and between 5 KB and 50 MB; clean single-speaker recordings clone best. The returned id works immediately in the speech operations.
  • List Cloned Voices Lists every cloned voice on the account with the manifest id the speech operations expect, the name, and the clone type. The lookup before generating with, or deleting, a clone.
  • Delete Cloned Voice Permanently deletes a cloned voice, freeing a slot against the plan's clone limit. Flows still referencing the deleted id fail on their next request, which is exactly why this call sits behind human confirmation.

Frequently Asked Questions

What can FlowRunner do with PlayHT?

FlowRunner agents can run Generate Speech, Stream Speech, and Get TTS Job in PlayHT, plus 4 more actions.

Does connecting PlayHT to FlowRunner require OAuth?

No. PlayHT connects to FlowRunner with an API key, no OAuth flow required.

Can PlayHT trigger a FlowRunner workflow automatically?

PlayHT doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.

Start building with PlayHT

$100 in credits. No card required. Connect in minutes.