AssemblyAI

Transcribe audio, analyze content, and interact with LLM gateway using AssemblyAI

Overview

The AssemblyAI tool provides comprehensive speech-to-text transcription, audio intelligence features (sentiment analysis, entity detection, auto chapters, content safety, PII redaction, summarization), and an LLM gateway for chat completions and understanding. It supports async transcript creation with built-in polling, file upload for audio binary data, subtitle export (SRT/VTT), redacted audio download, word search, and paragraph/sentence segmentation. Uses the AssemblyAI REST API v2 with API key authentication.

Category: Utility
Tool Name: assemblyai
Version: 1

Appearance: Icon: lucide-Headphones | Color: #2545D3

Node Type

Action (Binary) — handles file/binary data operations

Input / Output

Direction	Port(s)
Input	Input
Output	Output, Error

Credentials

This tool requires AssemblyAI API credentials. See the Credentials Guide for setup instructions.

Resources

Resource	Value
File	file
Transcript	transcript
LLM Gateway	llmGateway

Operations

Every operation belongs to one resource: File offers Upload, Transcript offers the nine transcript operations, and LLM Gateway offers Chat Completions and Understanding.

Operation	Value	Description
Upload	upload	Upload a binary audio file to AssemblyAI
Create	create	Create a new transcript from an audio URL
Get	get	Get a transcript by ID
List	list	List all transcripts
Delete	delete	Delete a transcript by ID
Get Paragraphs	getParagraphs	Get transcript split into paragraphs
Get Sentences	getSentences	Get transcript split into sentences
Get Subtitles	getSubtitles	Export transcript as SRT or VTT subtitles
Get Redacted Audio	getRedactedAudio	Download the redacted version of the audio
Word Search	wordSearch	Search for words in a transcript
Chat Completions	chatCompletions	Send chat messages through the LLM gateway
Understanding	understanding	Use the understanding endpoint for audio analysis

Parameters

File: Upload

Parameter	Type	Required	Default	Description
Binary Property	string	No	data	Name of the binary property containing the audio file to upload. Names are case-sensitive — see the upstream node’s Binary Data panel for the exact names to use. Supports expressions.

Transcript: Create

Parameter	Type	Required	Default	Description
Audio URL	string	Yes	—	URL of the audio file to transcribe. Can be a public URL or an AssemblyAI upload URL. Supports expressions.
Wait for Completion	boolean	No	true	Whether to poll and wait until the transcript is completed before returning.
Polling Interval (ms)	number	No	3000	Milliseconds between polling attempts when waiting for transcript completion. (shown when Wait for Completion is true)
Max Polling Attempts	number	No	60	Maximum number of polling attempts before timing out. (shown when Wait for Completion is true)
Transcript Options	collection	No	{}	Additional options for transcript creation.
— Speech Model	options	No	best	The speech recognition model to use.
				Options: best (highest accuracy model), nano (faster, lower-cost model)
— Language Code	string	No	en	The language code of the audio (e.g., en, es, fr, de).
— Language Detection	boolean	No	false	Whether to automatically detect the language of the audio.
— Speaker Labels	boolean	No	false	Whether to enable speaker diarization.
— Speakers Expected	number	No	0	Expected number of speakers. Set to 0 for automatic detection.
— Punctuate	boolean	No	true	Whether to add punctuation to the transcript.
— Format Text	boolean	No	true	Whether to format text (e.g., numbers, dates).
— Sentiment Analysis	boolean	No	false	Whether to enable sentiment analysis on each sentence.
— Entity Detection	boolean	No	false	Whether to detect entities (people, places, organizations, etc.).
— Auto Chapters	boolean	No	false	Whether to automatically generate chapters with summaries.
— Auto Highlights	boolean	No	false	Whether to automatically detect key phrases and highlights.
— Content Safety	boolean	No	false	Whether to detect sensitive content topics.
— Redact PII	boolean	No	false	Whether to redact personally identifiable information from the transcript.
— Redact PII Policies	string	No	—	Comma-separated PII policies to redact (e.g., medical_process, phone_number, email_address, credit_card_number, ssn, date_of_birth, person_name).
— Summarization	boolean	No	false	Whether to generate a summary of the transcript.
— Summary Model	options	No	informative	The summarization model to use.
				Options: informative (best for informational content), conversational (best for conversational audio), catchy (best for catchy, engaging summaries)
— Summary Type	options	No	bullets	The type of summary to generate.
				Options: bullets (concise bullet points), bullets_verbose (detailed bullet points), gist (brief one-line summary), headline (headline-style summary), paragraph (full paragraph summary)
— Webhook URL	string	No	—	URL to receive a webhook notification when transcription is complete.
— Webhook Auth Header Name	string	No	—	Name of the authentication header for the webhook.
— Webhook Auth Header Value	string	No	—	Value of the authentication header for the webhook.
— Dual Channel	boolean	No	false	Whether the audio is dual-channel (stereo) with separate speakers per channel.
— Filter Profanity	boolean	No	false	Whether to filter profanity from the transcript text.
— Word Boost	string	No	—	Comma-separated words or phrases to boost recognition accuracy for.

Transcript: Get

Parameter	Type	Required	Default	Description
Transcript ID	string	Yes	—	The ID of the transcript. Supports expressions.

Transcript: List

Parameter	Type	Required	Default	Description
Return All	boolean	No	false	Whether to return all transcripts using cursor pagination.
Limit	number	No	20	Maximum number of transcripts to return. (shown when Return All is false)

Transcript: Delete

Parameter	Type	Required	Default	Description
Transcript ID	string	Yes	—	The ID of the transcript. Supports expressions.

Transcript: Get Paragraphs

Parameter	Type	Required	Default	Description
Transcript ID	string	Yes	—	The ID of the transcript. Supports expressions.

Transcript: Get Sentences

Parameter	Type	Required	Default	Description
Transcript ID	string	Yes	—	The ID of the transcript. Supports expressions.

Transcript: Get Subtitles

Parameter	Type	Required	Default	Description
Transcript ID	string	Yes	—	The ID of the transcript. Supports expressions.
Subtitle Format	options	No	srt	The subtitle format to export.
				Options: srt, vtt

Transcript: Get Redacted Audio

Parameter	Type	Required	Default	Description
Transcript ID	string	Yes	—	The ID of the transcript. Supports expressions.
Output Binary Property	string	No	data	Name of the binary property to store the redacted audio file.
Output Filename	string	No	redacted_audio.wav	Filename for the downloaded redacted audio file.

Transcript: Word Search

Parameter	Type	Required	Default	Description
Transcript ID	string	Yes	—	The ID of the transcript. Supports expressions.
Search Words	string	Yes	—	Comma-separated list of words to search for in the transcript. Supports expressions.

LLM Gateway: Chat Completions

Parameter	Type	Required	Default	Description
Chat Messages	string	No	—	JSON array of chat messages (e.g., [{“role”: “user”, “content”: “Summarize this transcript”}]). Supports expressions.
Chat Model	string	No	anthropic/claude-3-5-sonnet	The LLM model to use for chat completions (e.g., anthropic/claude-3-5-sonnet).
Transcript IDs	string	No	—	Comma-separated transcript IDs to provide as context for the LLM chat. Supports expressions.

LLM Gateway: Understanding

Parameter	Type	Required	Default	Description
Understanding Input	string	No	—	JSON input for the understanding endpoint. Supports expressions.

All Operations

Parameter	Type	Required	Default	Description
Max Concurrency	number	No	5	Maximum number of items to process concurrently.

Output Data

One output item per input item. The AssemblyAI response is merged into the item JSON at the top level — its fields sit alongside the fields already on the item, and any field of the same name is overwritten. Binary data from the input item is forwarded unchanged; Get Redacted Audio additionally attaches the downloaded audio as a new binary property.

Operation	What lands on the item
Upload	The upload response, including the upload_url you hand to Create.
Create	The transcript object — id, status, the full transcript text with word timestamps, and whichever intelligence results you enabled (speaker labels, sentiment scores, detected entities, chapters, highlights, content safety, summary). With Wait for Completion off the node returns immediately and status is still queued or processing.
Get	The same transcript object, fetched by ID.
List	transcripts (the array of transcript records) plus the API’s own pagination details. With Return All on, every page is fetched and the item carries transcripts and total_count instead.
Delete	The deletion response for that transcript.
Get Paragraphs	The transcript split into paragraphs, as returned by AssemblyAI.
Get Sentences	The transcript split into sentences, as returned by AssemblyAI.
Get Subtitles	text — the whole SRT or VTT document as a single string — and format (srt or vtt).
Get Redacted Audio	redacted_audio_url and status, plus the downloaded audio file stored under the Output Binary Property.
Word Search	The word-search response for the words you supplied.
Chat Completions	The LLM gateway chat response.
Understanding	The understanding endpoint response.

Reference the result downstream by expression, e.g. {{ $json.upload_url }} after Upload or {{ $json.id }} after Create.

Usage Examples

• Transcribe an audio file from a URL with speaker labels
• Upload binary audio data and get a transcription URL
• Get transcript paragraphs and sentences for formatting
• Export subtitles in SRT or VTT format
• Download redacted audio with PII removed
• Search for specific words in a transcript
• Use LLM gateway for chat completions with transcript context
• Analyze audio for sentiment, entities, and content safety

Example Configuration

Upload an audio file:

{
  "resource": "file",
  "operation": "upload",
  "binaryPropertyName": "data",
  "maxConcurrency": 5
}

Create a transcript without waiting for it to finish:

{
  "resource": "transcript",
  "operation": "create",
  "audioUrl": "https://example.com/audio/interview.mp3",
  "waitForCompletion": false,
  "maxConcurrency": 5
}

Create a transcript and wait for the result:

{
  "resource": "transcript",
  "operation": "create",
  "audioUrl": "https://example.com/audio/interview.mp3",
  "waitForCompletion": true,
  "pollingInterval": 3000,
  "maxPollingAttempts": 60,
  "maxConcurrency": 5
}

Create a transcript with advanced options:

{
  "resource": "transcript",
  "operation": "create",
  "audioUrl": "https://storage.example.com/podcast.mp3",
  "waitForCompletion": true,
  "pollingInterval": 5000,
  "maxPollingAttempts": 30,
  "transcriptOptions": {
    "speech_model": "best",
    "language_code": "en",
    "language_detection": false,
    "speaker_labels": true,
    "speakers_expected": 2,
    "punctuate": true,
    "format_text": true,
    "sentiment_analysis": false,
    "entity_detection": true,
    "auto_chapters": false,
    "auto_highlights": true,
    "content_safety": false,
    "redact_pii": false,
    "redact_pii_policies": "",
    "summarization": false,
    "summary_model": "informative",
    "summary_type": "bullets",
    "webhook_url": "",
    "webhook_auth_header_name": "",
    "webhook_auth_header_value": "",
    "dual_channel": false,
    "filter_profanity": false,
    "word_boost": ""
  },
  "maxConcurrency": 5
}

Create a transcript with PII redaction and summarization:

{
  "resource": "transcript",
  "operation": "create",
  "audioUrl": "https://cdn.example.com/call-recording.wav",
  "waitForCompletion": true,
  "pollingInterval": 4000,
  "maxPollingAttempts": 50,
  "transcriptOptions": {
    "speech_model": "best",
    "language_code": "en",
    "language_detection": false,
    "speaker_labels": false,
    "speakers_expected": 0,
    "punctuate": true,
    "format_text": true,
    "sentiment_analysis": true,
    "entity_detection": false,
    "auto_chapters": false,
    "auto_highlights": false,
    "content_safety": false,
    "redact_pii": true,
    "redact_pii_policies": "person_name,phone_number,email_address",
    "summarization": true,
    "summary_model": "informative",
    "summary_type": "bullets",
    "webhook_url": "",
    "webhook_auth_header_name": "",
    "webhook_auth_header_value": "",
    "dual_channel": false,
    "filter_profanity": true,
    "word_boost": ""
  },
  "maxConcurrency": 3
}

Get a transcript by ID:

{
  "resource": "transcript",
  "operation": "get",
  "transcriptId": "abc123xyz",
  "maxConcurrency": 5
}

Delete a transcript:

{
  "resource": "transcript",
  "operation": "delete",
  "transcriptId": "abc123xyz",
  "maxConcurrency": 5
}

Get paragraphs:

{
  "resource": "transcript",
  "operation": "getParagraphs",
  "transcriptId": "abc123xyz",
  "maxConcurrency": 5
}

Get sentences:

{
  "resource": "transcript",
  "operation": "getSentences",
  "transcriptId": "abc123xyz",
  "maxConcurrency": 5
}

Export subtitles as SRT:

{
  "resource": "transcript",
  "operation": "getSubtitles",
  "transcriptId": "abc123xyz",
  "subtitleFormat": "srt",
  "maxConcurrency": 5
}

Export subtitles as VTT:

{
  "resource": "transcript",
  "operation": "getSubtitles",
  "transcriptId": "abc123xyz",
  "subtitleFormat": "vtt",
  "maxConcurrency": 5
}

Download redacted audio:

{
  "resource": "transcript",
  "operation": "getRedactedAudio",
  "transcriptId": "abc123xyz",
  "outputBinaryPropertyName": "redactedAudio",
  "outputFilename": "redacted_output.mp3",
  "maxConcurrency": 5
}

Search a transcript for words:

{
  "resource": "transcript",
  "operation": "wordSearch",
  "transcriptId": "abc123xyz",
  "searchWords": "contract,payment,deadline",
  "maxConcurrency": 5
}

List transcripts up to a limit:

{
  "resource": "transcript",
  "operation": "list",
  "returnAll": false,
  "limit": 25,
  "maxConcurrency": 5
}

List every transcript, paginating automatically:

{
  "resource": "transcript",
  "operation": "list",
  "returnAll": true,
  "maxConcurrency": 5
}

Leave limit out when Return All is on — it only applies when Return All is off.

Send chat messages through the LLM gateway with transcript context:

{
  "resource": "llmGateway",
  "operation": "chatCompletions",
  "chatMessages": "[{\"role\": \"user\", \"content\": \"Summarize the key decisions from this meeting transcript.\"}]",
  "chatModel": "anthropic/claude-3-5-sonnet",
  "chatTranscriptIds": "abc123xyz,def456uvw",
  "maxConcurrency": 5
}

Send chat messages without transcript context:

{
  "resource": "llmGateway",
  "operation": "chatCompletions",
  "chatMessages": "[{\"role\": \"system\", \"content\": \"You are a helpful assistant.\"}, {\"role\": \"user\", \"content\": \"What are best practices for meeting notes?\"}]",
  "chatModel": "openai/gpt-4o",
  "chatTranscriptIds": "",
  "maxConcurrency": 5
}

Call the understanding endpoint:

{
  "resource": "llmGateway",
  "operation": "understanding",
  "understandingInput": "{\"audio_url\": \"https://example.com/audio.mp3\", \"questions\": [{\"question\": \"What is the main topic?\", \"answer_format\": \"brief\"}]}",
  "maxConcurrency": 5
}

Upload, then transcribe (two nodes)

Upload a local binary audio file first, then use the returned upload URL to create the transcript.

Step 1 — upload node:

{
  "resource": "file",
  "operation": "upload",
  "binaryPropertyName": "data",
  "maxConcurrency": 5
}

Step 2 — create transcript node, reading upload_url from step 1:

{
  "resource": "transcript",
  "operation": "create",
  "audioUrl": "={{$json.upload_url}}",
  "waitForCompletion": true,
  "pollingInterval": 3000,
  "maxPollingAttempts": 60,
  "transcriptOptions": {
    "speech_model": "best",
    "language_code": "en",
    "language_detection": false,
    "speaker_labels": false,
    "speakers_expected": 0,
    "punctuate": true,
    "format_text": true,
    "sentiment_analysis": false,
    "entity_detection": false,
    "auto_chapters": false,
    "auto_highlights": false,
    "content_safety": false,
    "redact_pii": false,
    "redact_pii_policies": "",
    "summarization": false,
    "summary_model": "informative",
    "summary_type": "bullets",
    "webhook_url": "",
    "webhook_auth_header_name": "",
    "webhook_auth_header_value": "",
    "dual_channel": false,
    "filter_profanity": false,
    "word_boost": ""
  },
  "maxConcurrency": 5
}

Transcribe a public URL with speaker labels

{
  "resource": "transcript",
  "operation": "create",
  "audioUrl": "https://cdn.example.com/webinar-recording.mp4",
  "waitForCompletion": true,
  "pollingInterval": 5000,
  "maxPollingAttempts": 120,
  "transcriptOptions": {
    "speech_model": "best",
    "language_code": "en",
    "language_detection": false,
    "speaker_labels": true,
    "speakers_expected": 3,
    "punctuate": true,
    "format_text": true,
    "sentiment_analysis": false,
    "entity_detection": false,
    "auto_chapters": true,
    "auto_highlights": false,
    "content_safety": false,
    "redact_pii": false,
    "redact_pii_policies": "",
    "summarization": true,
    "summary_model": "informative",
    "summary_type": "bullets",
    "webhook_url": "",
    "webhook_auth_header_name": "",
    "webhook_auth_header_value": "",
    "dual_channel": false,
    "filter_profanity": false,
    "word_boost": "Q3,roadmap,milestone"
  },
  "maxConcurrency": 3
}

Transcribe, redact PII, then download the redacted audio

Step 1 — create the redacted transcript:

{
  "resource": "transcript",
  "operation": "create",
  "audioUrl": "https://example.com/sensitive-call.mp3",
  "waitForCompletion": true,
  "pollingInterval": 4000,
  "maxPollingAttempts": 60,
  "transcriptOptions": {
    "speech_model": "best",
    "language_code": "en",
    "language_detection": false,
    "speaker_labels": false,
    "speakers_expected": 0,
    "punctuate": true,
    "format_text": true,
    "sentiment_analysis": false,
    "entity_detection": false,
    "auto_chapters": false,
    "auto_highlights": false,
    "content_safety": false,
    "redact_pii": true,
    "redact_pii_policies": "person_name,phone_number,email_address,ssn",
    "summarization": false,
    "summary_model": "informative",
    "summary_type": "bullets",
    "webhook_url": "",
    "webhook_auth_header_name": "",
    "webhook_auth_header_value": "",
    "dual_channel": false,
    "filter_profanity": false,
    "word_boost": ""
  },
  "maxConcurrency": 5
}

Step 2 — download the redacted audio:

{
  "resource": "transcript",
  "operation": "getRedactedAudio",
  "transcriptId": "={{$json.id}}",
  "outputBinaryPropertyName": "redactedAudio",
  "outputFilename": "redacted_call.mp3",
  "maxConcurrency": 5
}

Analyze a transcript with the LLM gateway

Step 1 — get the transcript:

{
  "resource": "transcript",
  "operation": "get",
  "transcriptId": "abc123xyz",
  "maxConcurrency": 5
}

Step 2 — chat completions using that transcript as context:

{
  "resource": "llmGateway",
  "operation": "chatCompletions",
  "chatMessages": "[{\"role\": \"user\", \"content\": \"Extract all action items and assigned owners from this transcript.\"}]",
  "chatModel": "anthropic/claude-3-5-sonnet",
  "chatTranscriptIds": "abc123xyz",
  "maxConcurrency": 5
}

Search for keywords across a transcript

{
  "resource": "transcript",
  "operation": "wordSearch",
  "transcriptId": "abc123xyz",
  "searchWords": "budget,approval,Q4,launch",
  "maxConcurrency": 5
}

Error Handling

Mode	Behavior
stop	Halts workflow on first error
continue	Skips failed items, passes successful ones through
errorPort	Routes failed items to Error output port

Tips

AssemblyAI transcribes audio via speech-to-text and applies intelligence features including sentiment analysis, entity detection, PII redaction, content safety, and auto-chapter generation. Use it when a workflow needs to convert audio recordings into structured text or extract insights like sentiment, named entities, or topic summaries from spoken content. It outputs a transcript with full text, word timestamps, paragraph and sentence segments, detected entities, sentiment scores, and optional SRT or VTT subtitle files.

Key Rules and Gotchas

1. transcriptOptions is a flat collection object. Never wrap it in _default or an array. All sub-fields sit directly under the transcriptOptions key.

2. waitForCompletion defaults to true, so create blocks while it polls. A long recording can hold the node open for pollingInterval × maxPollingAttempts (3 minutes at the defaults of 3000 ms × 60). Set waitForCompletion: false for fire-and-forget creates, and raise maxPollingAttempts for long audio. pollingInterval and maxPollingAttempts only apply when waitForCompletion is true and operation is "create" — do not include them in other contexts.

3. limit is only valid when returnAll is false on the list operation. When returnAll is true, omit limit.

4. chatMessages must be a JSON-serialized string, not a raw JSON object. Escape inner quotes appropriately.

5. transcriptId is shared across multiple operations (get, delete, getParagraphs, getSentences, getSubtitles, getRedactedAudio, wordSearch) — always required for single-transcript operations other than create.

6. resource and operation must be consistent. For example, "upload" belongs to resource: "file", transcript CRUD operations belong to resource: "transcript", and "chatCompletions" / "understanding" belong to resource: "llmGateway".