tertiusLong recordings fail differently from short dictation. Files can have no readable audio track, unusual codecs, hours of silence, changing speakers, background noise, or terminology that appears only near the end. A useful workflow validates early, reports progress, preserves context between segments, and lets the user cancel.
How Tertius handles a media file
- Validate: confirm the file type, duration, and readable audio track before an expensive request begins.
- Prepare locally: extract a compact temporary audio copy on the Windows computer. The original audio or video remains in its folder.
- Transcribe in segments: send bounded parts to the transcription service while reporting progress and carrying context through the recording.
- Apply the playbook: turn the transcript into the selected outcome, such as meeting notes, actions, summary, or clean text.
- Clean up: remove temporary files and retain text history only when the user has enabled it.
Tertius does not claim that the original recording “never leaves your device” in every form. The source file is not uploaded as a whole, but temporary audio segments are sent through the authenticated Tertius service to OpenAI for transcription.
Choose the outcome before the upload
A raw transcript is often only an intermediate artifact. Selecting the outcome first helps define the structure without pretending the model can infer every business need.
Meeting notes
Organize decisions, open questions, risks, and next steps from a recorded call.
Action items
Extract concrete tasks and preserve owners or dates only when the speakers stated them.
Summary
Reduce a long recording to its central points while keeping the transcript available for reference.
Clean transcript
Improve readability by reducing filler and false starts without inventing new content.
What affects transcription quality
- Microphone distance, clipping, room echo, background noise, and overlapping speakers.
- Accents, language, speaking rate, proper names, acronyms, and domain terminology.
- Compression artifacts and whether the video contains a clear, consistent audio track.
- The amount of human review applied after automated transcription and outcome processing.
For important work, sample the beginning, middle, and end; verify names and numbers against the recording; and keep the source available during review. Tertius intentionally avoids publishing a universal accuracy percentage because no single number describes every recording.
Common questions
Which formats can I import?
Common formats include MP3, WAV, M4A, MP4, MOV, MKV, and WebM when the file contains a readable audio track.
What is the maximum recording length?
The private beta accepts a file up to three hours long.
Does Tertius upload the original video?
No. It extracts temporary audio locally, sends bounded audio segments for transcription, and deletes temporary files. The source file remains in its chosen folder.
Do I need an OpenAI account or ChatGPT subscription?
No. Sign in with your approved tester email and a six-digit code. Tertius keeps its OpenAI credential on the server and applies the account’s tester allowance.