AI-assisted transcription workflow: from raw audio to final text

Clarify the client request before you touch the audio

Before you upload anything to an AI tool, get crystal clear on what the client wants. Ask about audio length, number of speakers, language or accents, and deadline. Confirm whether they need verbatim or clean transcription and if they have a style guide or template. Clarify how they want timestamps and speaker labels. These details shape the settings you choose later and how much manual editing you plan. A short written summary of the agreement in email or chat helps prevent scope creep and disputes about what is “included” in the price. This planning step saves more time than any tool upgrade.

Prepare and organise your audio files first

AI tools handle clear, organised audio better. Download or copy the original files to a dedicated project folder on your computer or cloud drive. Rename files with a clear pattern, for example “ClientX_Interview1_2026-08-02.mp3”. If the recording is very quiet or has background noise, lightly clean it with basic audio software such as Audacity, but avoid heavy filters that distort speech. Split huge sessions into logical parts if your AI tool has upload limits. Make a simple text note with file names, duration and any known issues. Good prep reduces upload errors and helps you track time for billing later.

Generate an AI draft transcript with the right settings

Once your audio is ready, upload it to your chosen speech to text tool. Select the correct language and, if possible, the region or accent. Turn on speaker diarization if the platform supports it, then choose a sensible number of speakers. Use options for basic punctuation and casing. Do not rely on automatic formatting for specialist content such as legal or medical terms, plan to correct these yourself. When the draft is ready, export it in a flexible format such as TXT or DOCX, not only as a PDF. Keep the raw AI output unchanged in case you need to compare later while you edit.

Edit the AI draft with focused listening passes

AI gives you a starting point, your value comes from careful editing. Do one fast pass to fix obvious errors such as wrong names, numbers, and punctuation. Then do a slower pass with audio at normal speed. Use shortcuts to pause, rewind a few seconds, and insert timestamps where needed. Mark any unclear sections with tags like “inaudible” plus a timestamp so the client sees you tried. If the client requested clean transcription, remove filler words and false starts while you listen. Aim for consistent speaker labels and formatting across the whole file, because this is where many beginners lose quality.

Run final checks and format for delivery

Before sending anything, check the transcript against the client instructions. Confirm that timestamps follow the requested pattern and that speaker labels are consistent. Run a spell check, then manually review names, brands and technical terms, since software often misses these. Make sure the file name, document title and page headers match the project. Apply simple, clean formatting such as readable font, clear paragraph breaks and numbered questions if needed. Add a short note in your delivery message about any audio issues or “inaudible” sections, so the client understands the limits of the recording, not your effort.

Useful links

• Audacity audio editor — Free tool for basic noise reduction and volume adjustment.
• Google Docs — Useful for online editing and sharing transcripts with clients.

Keep learning