Dashboard
New Job
Groups this job's transcripts and prefixes the output filenames. Leave blank and we'll use the source name.
Embedded into Obsidian frontmatter, Logseq properties, RAG chunks, and the Anki tags column for filtering. NotebookLM & e-reader formats ignore them.
Supported file types
Telling pyannote the speaker count dramatically improves diarization on noisy audio. Choosing 1 speaker turns diarization off.
Names, jargon, topic, or who's speaking — improves WhisperX proper-noun accuracy and helps label speakers correctly (e.g. "Host Jane interviews guest John").
NotebookLM chunking and Anki card generation send short transcript excerpts to Anthropic (Claude) to build your output; WhisperX speaker labeling may send a brief excerpt too. See our Privacy Policy for how this data is handled.
Loading into LangChain / LlamaIndex
JSONLoader(path, jq_schema=".", content_key="text", json_lines=True)
# LlamaIndex
TextNode(text=rec["text"], metadata={k: v for k, v in rec.items() if k != "text"})
With an Anthropic key, cards are generated by Claude; otherwise one summary card per episode.
Sign in to run jobs
Save your transcripts and come back to them anytime.
Keyboard shortcuts
Single-key shortcuts are suppressed while typing in a text field.