Record a Podcast¶
Goal¶
Record each podcast microphone as a separate audio track, then get named transcripts for individual speakers and a combined transcript for reviewing the episode. Optionally create a draft summary or show notes with an LLM post-processor.
What You Need¶
- One physical audio interface with an input channel for each microphone. Aggregate and virtual devices are not supported.
- Microphones connected to separate inputs on that interface.
- A ready transcription model and enough space for lossless audio and temporary transcription chunks.
This workflow captures channels from that interface. It does not provide remote guest calling or mix in audio from other applications. For an episode already recorded elsewhere, use Transcribe Audio and Video Files.
Set Up the Microphones¶
- Open
Recording Settings...and find Multitrack Inputs. - Use Add Track for each microphone. Name tracks for their speakers, such as
HostandGuest, and enable the tracks you want to record. - Select your physical interface. Automatic input assignments follow track order, including disabled tracks; use each track's Input menu if your cabling differs.
- Use Check Input Levels and have each person speak. Confirm that the expected track responds and resolve duplicate input assignments before starting.
- Stop the input check when finished. Saved names and mappings remain available for your next episode, including when the interface is disconnected.
An unavailable input is skipped, so check that every intended microphone is available before recording. See the multitrack guide for mapping and reconnect behaviour.
Choose Your Outputs¶
| Setting | Starting choice | What you get |
|---|---|---|
| Per-track transcripts | On | A transcript for each microphone |
| Combined transcript | On | One episode transcript with track labels |
| TXT output | On | Readable text for editing and review |
| SRT or WebVTT | Only if needed | Captions for your editor or player |
| Word timing sidecar | Optional | Word-level timing for downstream tools |
| Chunk transcription | On for processing during capture; Off to defer it | Final exports are assembled after Stop in either mode |
| Reduce quiet microphone bleed for transcription | Off initially | Enable and test if other microphones pick up quiet spill |
| LLM post-processor output | Optional | A companion summary of the combined transcript |
| Also run the enabled LLM post-processor on each track | Off initially | Enable only when you want separate requests for individual tracks |
Track names supply speaker labels without a diarisation model. Original audio is saved as lossless PCM CAF at the interface's sample rate; the ordinary recording bitrate and sample-rate settings do not apply. Bleed reduction affects recognition audio only. It does not alter the saved originals or remove repeated words across microphones.
Record and Review an Episode¶
- Make a short test recording with every speaker, including a little overlapping speech, before a full episode.
- Choose Start Multitrack Recording... from Home or the menu bar, or start from Recording Settings.
- Keep the interface connected throughout capture. Track names and routing are locked while recording.
- Use Stop Recording when finished and keep OpenScreech running until finalisation completes.
- Open the session folder in your recording output directory. Combined exports are at its root; individual audio and enabled transcripts are inside numbered
tracks/folders. - Play each
recording.caf, check speaker labels intranscript.speakers.txt, and review the combined transcript against the audio. Check captions in your intended editor if enabled.
Overlapping speech retains track labels in the combined outputs. Words captured by more than one microphone are retained, so review microphone spill before publishing a transcript. If capture was interrupted, inspect session.json for errors and verify the available audio.
Optional: Draft Show Notes¶
Select a dedicated processor in Recording Settings → LLM Post-Processor, keep Processor Input set to Original Transcript, and enable LLM output. The multitrack original transcript already includes track labels.
Try this prompt in the processor tester:
Draft show notes from this podcast transcript.
Include a short episode summary and the main topics discussed.
Preserve speaker names and technical terms.
Do not invent facts, links, quotations, or timestamps.
Return only the draft show notes.
Review transcript.llm.txt against the transcript before publishing. A remote processor receives the transcript; choose a local processor when you want processing to remain on the Mac. Duplicate a shared processor before changing its prompt for podcast use.