Speechmatics
AITranscribe pre-recorded audio and video with Speechmatics in 60+ languages. Agents submit batch jobs from uploads or URLs, add speaker diarization, translation, summarization, and sentiment, then export transcripts as JSON, text, or SRT subtitles into FlowRunner file storage.
What This Integration Enables
The most operationally valuable data a company generates may be the least accessible: it is spoken. Sales calls, support conversations, interviews, all-hands recordings, hours of ground truth in a format nothing downstream can read. Speechmatics treats that backlog as an infrastructure problem: batch transcription in 60+ languages with speaker diarization, translation into up to five languages per pass, summarization, sentiment, topic detection, and subtitle export. This connector makes it a FlowRunner pipeline stage, and adds the operational follow-through the API's own design demands: Speechmatics retains jobs and transcripts for 7 days, so a transcription without a collection step is work scheduled for deletion. FlowRunner agents always finish the loop. - Submit uploads under 1 GB directly, or have Speechmatics fetch cloud-hosted files of any size from presigned URLs - Collapse submit, poll, and download into one step with server-side long polling for shorter recordings - Export json-v2 with word timings and confidence, plain text, or SRT subtitles, inline or into file storage - Choose the standard, enhanced, or melia-1 multilingual model, which detects and switches languages automatically - Audit account consumption by jobs and audio duration for cost control
Without FlowRunner
With FlowRunner
Use Case Scenarios
The call library that answers questions
Every recording that lands in [Dropbox](/integrations/dropbox-service) flows through Transcribe and Wait, and the diarized transcript posts to [Slack](/integrations/slack) with the summary up top. The json-v2 archive builds a searchable record of who said what and when, so "did we promise them the discount?" is answered with a timestamped quote instead of a memory contest.
One recording, five languages
A global support team runs QA on calls taken in whatever language the customer spoke. The melia-1 multilingual model transcribes recordings that switch languages mid-conversation, and the translation option renders each transcript into the review team's working languages in the same pass. Summaries and sentiment file into [Notion](/integrations/notion) per region, and the QA leads read one queue instead of five.
The archive backfill that respects its budget
Years of webinar recordings sit in [S3](/integrations/s3), many over the 1 GB direct-upload limit. The backfill flow generates presigned URLs and hands them to Submit Job from URL, so the bytes move straight from S3 to Speechmatics without transiting FlowRunner. Save Transcript to File writes SRT subtitles next to each video and the JSON into the knowledge pipeline. Get Usage reports consumed audio duration as the backfill proceeds, keeping the spend visible batch by batch rather than discovered on the invoice.
Human-in-Loop Highlight
Delete Job permanently destroys a job and everything attached to it: the stored transcript and the uploaded audio, gone before the 7-day auto-deletion would have taken them. In a pipeline handling sensitive recordings, medical dictation, legal interviews, HR conversations, that action is not cleanup, it is policy execution, and it cuts both ways. Purge too early and the transcript may not have been collected: the work is lost and the job still counts against usage. Purge too late and sensitive audio lingered on a third-party service longer than the company promised. So the FlowRunner purge flow makes the order of operations a contract: Save Transcript to File archives the output, the agent verifies the stored file exists and opens, and then a person confirms the purge list before Delete Job runs on each item. The confirmation names what is being destroyed and shows where its archive lives. Transcription runs at machine speed; the destruction of recordings runs at the speed of a signature.
Agent Capabilities
9 actionsTranscription
3- Submit Transcription Job Uploads an audio or video file (wav, mp3, aac, ogg, mpeg, amr, m4a, mp4, flac; under 1 GB) and starts an asynchronous batch job, with options for speaker or channel diarization, custom vocabulary, entity metadata, translation into up to five languages, summarization, sentiment, topics, and automatic chapters. Returns a job ID immediately.
- Submit Job from URL Starts a job for audio Speechmatics fetches itself from a publicly reachable URL, with optional authorization headers forwarded. The required path for files over the 1 GB direct-upload limit, and the natural fit for S3 or Azure Blob presigned URLs. Fetch failures end the job as rejected with per-attempt errors.
- Transcribe and Wait Submits a file and blocks until the transcript is ready, long-polling server-side with a cap of 840 seconds. Returns the finished text in one step, or the job ID with timedOut set when the cap is hit so the result can be collected later.
Transcripts
2- Get Transcript Downloads a completed job's transcript as json-v2 with per-word timings, confidence, speakers, entities, translations, summaries, sentiment, topics, and chapters; as plain text; or as an SRT subtitle file body. Every format also carries a plain-text transcript field.
- Save Transcript to File Downloads a completed transcript into FlowRunner file storage and returns a durable URL alongside the text, with the extension following the format: .json, .txt, or .srt. The archival step that outlives the 7-day retention window.
Jobs
3- Get Job Retrieves a job's status (running, done, rejected, or deleted), its configuration, source file name, audio duration, and any fetch or processing errors, with an optional server-side wait for a terminal state.
- List Jobs Lists the account's jobs, most recent first, up to 100 per request, paging backward with Created Before. Deleted jobs appear only when Include Deleted is enabled.
- Delete Job Permanently deletes a job with its stored transcript and uploaded audio, ahead of the automatic 7-day deletion. Cannot be undone, does not reduce reported usage, and runs only behind this page's human gate.
Account
1- Get Usage Reports consumed transcription as successfully processed jobs and total audio duration, broken down by processing mode and model, on UTC calendar days. The cost-control readout for backfills and steady-state pipelines alike.
Frequently Asked Questions
What can FlowRunner do with Speechmatics?
FlowRunner agents can run Submit Transcription Job, Submit Job from URL, and Transcribe and Wait in Speechmatics, plus 6 more actions.
Does connecting Speechmatics to FlowRunner require OAuth?
No. Speechmatics connects to FlowRunner with an API key, no OAuth flow required.
Can Speechmatics trigger a FlowRunner workflow automatically?
Speechmatics doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Speechmatics
$100 in credits. No card required. Connect in minutes.