transcription

Transcription

Transcribe Plaud recordings using ElevenLabs Scribe v2 with speaker diarization, or trigger Plaud's built-in (free, lower quality) processing.

COST WARNING

transcribe() calls ElevenLabs API and costs real credits. Before transcribing:

1. Check cache first: get_recording(file_ref) -- if transcript_cached: true, the call is FREE (returns from cache)
2. If not cached: warn the user before proceeding
3. For bulk: use memory_ingest(dry_run=true) to estimate cost first (see bulk-operations skill)

Tools

transcribe

High-quality transcription via ElevenLabs Scribe v2. Synchronous -- returns full transcript in one call.

Parameter	Type	Default	Description
file_ref	string	null	Plaud recording ID, prefix, or row number
file_path	string	null	Local audio file path (MP3, WAV, M4A, AAC, OGG, FLAC)
language	string	"eng"	3-letter ISO 639-3 code

Provide file_ref OR file_path, never both.

Response: { "status": "complete", "text": "...", "speakers": [...], "language": "eng" }

Timing: Cached transcripts return instantly. Uncached recordings take 60-120s (downloads audio, calls ElevenLabs, caches result). No polling needed.

trigger_processing

Trigger Plaud's built-in server-side transcription + summarization. Free but lower quality.

Parameter	Type	Default	Description
file_ref	string	required	File ID, prefix, or row number
language	string	"en"	2-letter Plaud code (NOT 3-letter ElevenLabs)
diarization	bool	true	Enable speaker identification

Idempotent -- safe to call on already-processed files. Processing takes 2-5 minutes on Plaud servers.

get_processing_status

Check Plaud's AI processing queue. No parameters. Returns list of files currently being processed with status.

list_languages

Show all supported transcription languages with codes. No parameters.

ElevenLabs vs Plaud Comparison

	ElevenLabs (transcribe)	Plaud (trigger_processing)
Quality	High accuracy, speaker diarization	Basic
Cost	ElevenLabs API credits	Free
Speed	~30s-2min per recording	2-5 minutes
Language codes	3-letter (eng, fra, deu)	2-letter (en, fr, de)
Caching	RustFS local cache	Plaud servers
Local files	Supported (file_path)	Not supported

Common ElevenLabs language codes: eng (English), fra (French), deu (German), jpn (Japanese), zho (Chinese), spa (Spanish), kor (Korean)

Pipeline

Plaud S3 --> RustFS audio cache --> ElevenLabs Scribe v2 --> RustFS transcript cache --> Cognee ingestion

Workflows

"Transcribe recording X"

1. get_recording(file_ref="<id>") -- check transcript_cached
2. If cached: transcribe(file_ref="<id>") -- instant, free
3. If not cached: warn user about cost, then transcribe(file_ref="<id>")
4. Format speaker segments with timestamps

"Transcribe a local audio file"

1. transcribe(file_path="/path/to/recording.mp3")
2. No Plaud account needed -- uses ElevenLabs directly

"Process recording with Plaud (free)"

1. trigger_processing(file_ref="<id>", language="en")
2. get_processing_status() -- confirm queued
3. Tell user it takes 2-5 minutes
4. Use get_content(file_ref, content_type="summary") after processing completes

"What languages are supported?"

1. list_languages() -- returns full list with codes for both systems

Anti-Patterns (NEVER DO)

• Never call transcribe() without checking cache first -- always get_recording first to avoid unnecessary cost
• Never mix language code formats -- ElevenLabs uses 3-letter (eng), Plaud uses 2-letter (en)
• Never use trigger_processing for local files -- it only works with Plaud recordings
• Never bulk-transcribe without dry_run -- use the bulk-operations skill with dry_run=true first