Cleanvoice
Media & VideoCleanvoice AI is an automated audio and podcast editor that removes filler words, silences, stutters, and mouth sounds, then normalizes loudness. Agents submit audio from a URL, apply a saved editing template, and pick up the cleaned file, transcript, and chapter summaries when the edit completes.
What This Integration Enables
Cleanvoice is not a transcription service that happens to clean audio. It is an editor, and the distinction matters, because most of what it does is subtraction: filler words removed, stutters removed, long silences trimmed, breathing softened or taken out. The output is shorter than the input and it says something slightly different from what was recorded. FlowRunner agents submit audio from a URL, apply a saved configuration so a show's sound is a setting rather than a habit, poll for completion on real booleans instead of a status string, and collect the cleaned file along with a transcript, chapter summaries, and social copy generated in the same pass.
Two behaviors shape any serious workflow around it. Reduce Noise runs against a different route from Create Edit and leaves the timing of the recording completely untouched, which makes it the safe choice when a file has to line up against an existing video or subtitle track. And Cleanvoice deletes edit files after a fixed retention window, so a flow that needs the output has to collect it, and a flow handling confidential material can call Delete Edit to remove the copy immediately rather than waiting for the window to close on its own.
Without FlowRunner
With FlowRunner
Use Case Scenarios
A show's sound as a setting, not a habit
Each client show has a saved configuration in Cleanvoice. When a raw file appears in Dropbox, the agent reads which show it belongs to, calls Get Configuration to confirm which flags that template actually applies, and submits Create Edit with the template id. On completion it collects the cleaned file, the transcript, and the chapter summaries, files them against the project in Asana, and posts the summary to the producer in Slack. Two editors working on two different shows no longer produce two different sounds, because neither of them is choosing the settings.
Video that still lines up
An interview was recorded on camera. The agent submits it with the video toggle set, because without it the video track is discarded and only the audio survives, and with the option that replaces cuts with silence so the output keeps the same duration as the input. The cleaned audio then lays straight back against the existing timeline and the existing subtitle file with no re-sync. The finished asset goes to Gumlet for delivery and the transcript goes into the show notes draft.
A recording that is a record
A client call is recorded for the account team's own reference. The intake agent recognizes from the tag that this file is not an episode, and it stops before choosing a treatment. The two options available do very different things to a file that may later be quoted, and the choice belongs to the person who will be quoting it.
Human-in-Loop Highlight
Create Edit and Reduce Noise produce files that sound like each other and are not the same artifact. Create Edit removes filler words, stutters, and long silences, so the timeline shortens and the recording no longer matches what was said. Reduce Noise runs the enhancement passes only, so every word and every pause survives. On a podcast episode that difference is a preference. On a client call, a deposition, or an incident debrief it is the difference between an edit and an alteration of a record. So the intake agent stops and asks the account lead: "This file is tagged as a client call, not an episode. Create Edit with filler word removal and silence trimming will cut audio and shorten the timeline. Reduce Noise leaves every word and every pause intact. Which treatment, and should the job be deleted from Cleanvoice as soon as the output is collected?" The lead answers in the channel. The human-in-the-loop step here costs one message and prevents an edit that no downstream reviewer would ever notice had happened.
Agent Capabilities
10 actionsAccount
1- Get Account Returns the account record and the remaining credit. The connection test, and the call that distinguishes a wrong key from an unreachable service, because every other route reports the same message for an absent, wrong, or revoked key.
Edits
4- Create Edit Submits audio or video for a full edit. Every cleaning and enhancement flag is a named parameter, and a saved configuration can supply them instead. Turning on summaries turns on transcription, and turning on social content turns on summaries, so the underlying options do not need to be set by hand.
- Reduce Noise Runs the enhancement passes only, against a separate route. It cleans the sound without removing words or trimming silences, so the timing of the recording is unchanged and the output still matches an existing video timeline or subtitle file.
- Get Edit Retrieves a job. The vendor returns a success response for every stage of processing with the real state in the body, so this operation adds explicit done and failed booleans that a flow can branch on without a string comparison a new stage would break.
- Delete Edit Removes a job and its files immediately. The operation a workflow handling confidential recordings calls as soon as it has collected the output, rather than waiting for the automatic retention window.
Uploads
2- Create Upload URL Returns a pre-signed URL to put file bytes to. The lower level half of the upload path, for flows that move the bytes themselves.
- Upload File From URL Runs the whole upload exchange in one call and returns the finished file URL. It reads the source as binary so media is not corrupted in transit, and it forwards no credentials, so the source has to be reachable without them.
Configurations
3- List Configurations Lists the saved editing templates on the account with their ids, names, and upload types.
- Get Configuration Returns a template's full settings object. The only way to see which flags a template actually applies before a workflow commits a file to it.
- List Configuration Tasks Lists the recently completed jobs submitted under one configuration. The audit view behind the trigger, and the way a flow reconciles what was processed against what was expected.
Triggers
1 triggersEvent Triggers
1- On Edit Completed Fires when a job submitted under one saved configuration finishes. Cleanvoice publishes no webhook, so this trigger polls the completed tasks list, and two limits follow from what that list contains: a job created without a configuration id never appears and never fires, and the list covers only the recent retention window. Its first pass records the current job ids as a baseline and emits nothing.
Frequently Asked Questions
What can FlowRunner do with Cleanvoice?
FlowRunner agents can run Get Account, Create Edit, and Reduce Noise in Cleanvoice, plus 7 more actions.
Does connecting Cleanvoice to FlowRunner require OAuth?
No. Cleanvoice connects to FlowRunner with an API key, no OAuth flow required.
Can Cleanvoice trigger a FlowRunner workflow automatically?
Yes. Cleanvoice supports 1 trigger that can start a FlowRunner workflow automatically.
Start building with Cleanvoice
$100 in credits. No card required. Connect in minutes.