Start with the episode you will publish
Save the episode title, final audio filename, language and revision date beside the transcript. Transcribing an earlier cut makes references unreliable once an introduction or advertisement is inserted. Keep an untouched draft, then make a separate reviewed copy. A transcript records the episode; show notes summarize it and should remain a separate document.
Choose an editing policy before correcting words
For a verbatim record, retain meaningful repetitions and unfinished phrases. For a lightly edited reading copy, remove distracting fillers only when meaning and emphasis remain intact, and label the copy as edited. Do not silently rewrite a guest’s argument. Confirm personal names and technical terms against the speaker’s own materials; replay numbers and dates. Search every occurrence after making a terminology correction.
Identify voices without inventing identities
Start a paragraph at each speaker change. Use a confirmed name or a neutral label such as Speaker 1 until the identity is known. Listen again to interruptions and overlapping speech. If a word cannot be recovered, use a consistent uncertainty note with its time reference rather than filling in a plausible sentence. Include relevant non-speech information when it is needed to understand the exchange; do not label every incidental noise.
Separate readable paragraphs from subtitle cues
A reading copy can combine adjacent cues into a sentence. Keep the underlying timed file separate: its start and end times still need to match playback. Test the beginning, a middle passage, the ending and every edited join. Split captions at meaningful phrase boundaries; check reading time on the intended player instead of applying one line-length rule to every language. Keep names and number-unit pairs together where practical.
Package and check the final delivery
Provide the reviewed reading copy, the matching audio revision and, if required, a tested SRT or VTT. TXT does not by itself create clickable navigation; that is a player feature. Translation is a second review task after source correction, not a substitute for transcription. Keep the source and translation distinguishable. Confirm the receiving platform’s current format and language requirements, then open the exported file and compare quoted passages with the audio once more.
Workflow sample
Public workflow sample: 33.72 seconds of self-authored English synthetic speech, processed on 23 October 2026. The linked SRT is the unmodified source-language output. It was checked in the recorded test, not professionally human-edited. This is not a full podcast, a multi-speaker test or a benchmark for this page’s language.
Original recording (English) · Original SRT (English)
3
00:00:07,950 --> 00:00:12,149
Our next workshop starts on June 15 at 9:30. The sample’s third cue runs from 00:00:07,950 to 00:00:12,149 and includes “June 15 at 9:30”. Preserve both values when editing. Joining cues into a reading paragraph must not overwrite the original cue times.
Frequently asked questions
Must every paragraph have a timestamp?
Not for ordinary reading. Add useful navigation references consistently; a timed subtitle file still needs valid cue timing.
Can I use the transcript as show notes?
Use it as evidence for separately written notes. A summary is not the full transcript, and quotations should remain faithful to the recording.
