Folder import finds common audio formats; multiple-file selection accepts other formats too. Each file must be under 1 GB. Files process one at a time using the settings below. Start fresh clears the current file and pending queue; completed transcripts stay in the list.
Audio and model options Audio language Auto detect English Arabic Chinese Spanish French German Hindi Italian Japanese Korean Portuguese Russian Turkish Persian Urdu Indonesian Dutch Polish Ukrainian Vietnamese Thai Hebrew Swedish Greek Romanian Tamil Bengali Malay Swahili
Whole minutes, 1–30. Longer chunks use API parts of up to 3 minutes to keep transcripts complete. Each part overlaps by 2 seconds.
Transcription model GPT Transcribe — ~$0.0045/min GPT-4o Mini Transcribe — ~$0.003/min GPT-4o Transcribe — ~$0.006/min Recommended · latest model ~$0.270 / hour
None Keep the original words from transcription. Join chunks locally with no AI cleanup; overlap repetitions may remain. Best for preserving raw evidence or editing yourself. Token usage: 0 cleanup tokens. Optional content titles add a small naming request.Minimal Repair chunk joins only: remove overlap repetitions and fix obvious cut words. Keep the rest of the wording and paragraphs unchanged. Best for clean recordings that need little editing. Token usage: Only about 600 characters per side of each join, plus instructions. A single chunk needs no cleanup request; optional content titles may need a small naming request. Many short chunks can cost more tokens than Standard.Standard Stitch edges, fix obvious recognition and punctuation errors, and create natural paragraphs. Best for everyday talks, interviews and meetings. Token usage: About 1× transcript input + 1× output, plus instructions and neighboring context.Advanced First identify topics, keywords and supported technical terminology across the recording, then edit with that guide. Best for lectures, research and specialist content. Preserve uncertain terms rather than guessing. Token usage: About 2× transcript input + 1× output, plus terminology guides, prompts, neighboring context and any reasoning.“1× transcript” means its token length. These are starting estimates; language, chunk count and model affect usage. Actual reported input/output tokens appear in the result. Transcription charges are separate.
On by default. Uses the selected model and existing processing calls when possible. Otherwise, one short request uses at most about 1,200 transcript characters sampled across the recording. Titles may miss details; the transcript itself stays unchanged.
Post-processing model GPT-6 Luna — default · fast & light GPT-6.1 Sol GPT-6 Astra $0.10 input / $0.50 output per 1M tokens
Verify with OpenAI before starting. This check uses no API credit.
Off by default. Your browser may ask to allow multiple downloads. Completed copies remain in the list either way.
Starting sends audio parts and transcript text directly to OpenAI and uses your API credit. Keep this tab open and your phone awake.