AI & transcription
Transcript, chapters and episode notes produced in a single analysis. Echo removal and speaker separation, with the original always kept.
The situation
Once the episode is edited, the work that makes it readable and findable remains: a transcript, chapters, episode notes. Done by hand, it takes a full second listen. A take recorded in a room that echoes, or two voices recorded on one microphone, usually get fixed in yet another application.
Episode post-production
On the rendered file, a single analysis produces the whole set:
- the full transcript, punctuated and split into paragraphs;
- the topics covered, turned into chapters;
- the paragraphs, turned into episode notes;
- each passage attributed to its speaker.
The language is the one set at launch, or the one the analysis detects. The result appears on the episode's publishing page. The transcript and chapters then go into the RSS feed, for the listening apps that display them, and the transcript also exists as a subtitle file.
Echo removal
In the editor, a source recorded in a room that echoes can be cleaned of that echo. The treatment runs on Podkstr's own servers. The result becomes a new version of the source: the original stays available and can be restored at any time. A source already cleaned does not go through cleaning again, which would degrade it.
Speaker separation
When several voices share one file, for instance two people on a single microphone or an imported recording, separation produces one source per speaker. Each voice is then edited on its own track, with its own effects.
Credits
Each treatment uses one credit: a post-production, an echo removal, a separation. The credit is counted at launch, and a refused request, such as a second cleaning of the same source, costs nothing. Nothing starts without an action from the team: an episode can be produced entirely without AI.
The number of credits depends on the plan, and full episode post-production is part of the higher plans.
Still missing
No episode summary is generated. Filler words are not removed automatically, and background noise is handled at the edit with the noise gate and the filter.
The other steps of an episode: live studio, multitrack editor, hosting. For a show run alone: solo creator.