How to Fix AI
Transcription Errors

1. 1

   Check the language and task

   If you wanted the original language but received English, check whether translation was enabled. OpenAI's Whisper API distinguishes transcription from translation into English.

2. 2

   Correct the existing draft

   Compare the text with the recording. Make corrections in your app's editor, or export the transcript and use an appropriate text or subtitle editor.

3. 3

   Supply terminology when supported

   If you rerun transcription, provide a short glossary of unusual names and acronyms where your app supports it. Whisper prompting has limits and does not work like unrestricted chatbot instructions.

4. 4

   Review the new result

   Check unclear passages, missing words, repeated text, and audio chunk boundaries. A newly generated transcript still needs verification against the recording.

For a few isolated mistakes, correcting the existing draft is usually more direct than repeatedly generating the entire transcript. If you export subtitles, check their timing against the matching recording.

OpenAI's speech to text documentation

1. 1

   Match the transcript to the recording

   Keep the original audio and an unchanged copy of the draft. Confirm that both cover the same recording.

2. 2

   Clarify the transcription request

   If generating a replacement draft in the Gemini web app, select Add files, then Upload Files. Request transcription in the original language, with unclear speech marked and summaries kept separate.

3. 3

   Edit while listening

   Copy the transcript into an editable document. Check names, numbers, missing words, and speaker changes against the audio. Verify suggested corrections before accepting them.

4. 4

   Check completeness and timestamps

   Review the beginning, middle, and end for omitted or repeated passages. Check speaker labels and timestamps manually. For subtitles, verify timing against the final video.

Upload limits vary by plan. When splitting a recording, preserve the original time offsets and check the joins for missing or duplicated speech.

File uploads Audio transcription capabilities

1. 1

   Find your cloud recording

   Sign in to the Zoom web portal. Open Recordings & Transcripts, then select Cloud recordings.

2. 2

   Open the transcript

   Choose the meeting and open its recording thumbnail. Locate the transcript panel beside the recording.

3. 3

   Correct the wording

   Replay the relevant passage. Hover over the incorrect phrase, select the pencil icon, and enter the words you hear.

4. 4

   Save and check speakers

   Select the check mark to save your correction. Review unknown or incorrect speaker names and confirm who said each passage.

Pay particular attention to decisions and action items. Correct words can still misrepresent a meeting when assigned to the wrong speaker.

Zoom's audio transcription instructions

1. 1

   Locate the recording

   One route is OneDrive → My files → Recordings. Find the recording that matches your transcript.

2. 2

   Open it in your browser

   Use the recording's menu to open it in the Microsoft 365 video player.

3. 3

   Correct the transcript

   Open Transcript. Hover over the relevant section, select Edit, and correct the wording while checking the recording.

4. 4

   Save the correction

   Select Done. Continue reviewing the recording, paying attention to names, numbers, and speaker changes.

A saved transcript does not necessarily mean a recording is available. Without the original audio, you can improve spelling and formatting, but cannot verify every spoken word.

Editing instructions Download instructions

1. 1

   Keep the recording and draft

   Save the original audio and transcript. Open an editable copy in your application or a document editor.

2. 2

   Correct words against the audio

   Listen for misheard names, numbers, technical terms, and missing speech. Mark unresolved passages with timestamps.

3. 3

   Improve a replacement draft

   If regenerating with Voxtral Mini Transcribe V2, use vocabulary guidance for names and specialist terms where available. Compare the new output with the recording before accepting changes.

4. 4

   Verify speakers and timing

   Check each speaker change and any generated timestamps. Review overlapping speech carefully. Check subtitle timing against the final video before delivery.

Mistral notes that Voxtral Mini Transcribe V2 typically transcribes one speaker during overlapping speech. Review these passages for omitted words from other speakers.

Transcription documentation Model features and limitations

1. 1

   Open the editor

   Select the completed file from your dashboard. Choose Edit Transcript in the right-hand column.

2. 2

   Review the wording

   Listen to the recording and correct misheard words, names, and numbers. Check speaker changes and look for missing or repeated speech.

3. 3

   Check the language settings

   If the output is in the wrong language, check Audio Language before transcribing again. Transcribe to English translates non-English speech into English; leave it off when you need the original language.

4. 4

   Export the corrected transcript

   Choose DOCX or TXT for an editable document, or SRT or VTT when you need subtitles. Check the exported file before sharing it.

If you try audio restoration or another transcription mode, compare the new result with the original recording before accepting changes.

Editing and subtitle guidance Available export formats

1. 1

   Select the incorrect text

   Open your recording in the editor. Select the word or phrase you want to correct.

2. 2

   Enter the replacement

   Choose Correct, type the wording you hear in the recording, and confirm the change.

3. 3

   Review repeated mistakes

   Use Correct all only when the same replacement is appropriate throughout the transcript. Matching is sensitive to capitalization and punctuation.

4. 4

   Check the captions

   Replay the passage and review any captions that use the corrected text.

If you shorten a podcast or video after proofreading, check the transcript and subtitle timing against the new edit.

Text corrections Repeated corrections Edited transcript downloads

1. 1

   Listen and correct

   Replay each passage and correct misheard words, names, numbers, and punctuation.

2. 2

   Use confidence highlighting

   Enable the heatmap with the thermometer icon. Lighter words indicate lower confidence and may deserve extra attention. They are not necessarily incorrect, and unflagged words still need checking.

3. 3

   Check speaker assignments

   Confirm who is speaking throughout the recording. If one person has multiple labels, review the cleanup options under File → Speakers Breakdown → Fix Speakers. Confirm identities before merging paragraphs.

4. 4

   Review the complete transcript

   Listen through the recording while following the corrected text. Check omissions and speaker changes as well as highlighted words.

The heatmap helps prioritize review. It does not automatically correct mistakes or certify the transcript's accuracy.

Confidence heatmap Speaker corrections Export instructions

1. 1

   Keep the original audio

   Retain a separate recording. ChatGPT Record deletes audio after transcription, so do not rely on it as your audio archive.

2. 2

   Prepare an editable transcript

   Save a copy of the full transcript in a document editor. Keep the original draft for comparison. A generated summary cannot replace the transcript.

3. 3

   Listen and correct

   Compare the transcript with your recording. Check names, numbers, missing words, and speaker labels. Mark unclear passages with timestamps instead of guessing.

4. 4

   Review related notes

   Check summaries and action items separately after correcting the transcript. Verify any quotations directly against the recording.

Without the original audio, you can edit the text but cannot fully verify what was spoken.

ChatGPT Record documentation

1. 1

   Open the meeting

   Select Meetings and choose the conversation you want to review.

2. 2

   Edit the transcript

   Open the Transcript tab and select the pencil icon. Compare the text with the recording and correct the wording.

3. 3

   Check speaker labels

   When renaming a speaker, choose whether the change applies to the current instance or throughout the transcript. If a paragraph contains multiple speakers, split it and assign the correct names.

4. 4

   Save your corrections

   Select Save, then review the corrected passages against the recording.

After correcting names or decisions, check the related summary and action items separately before sharing them.

Transcript editing Speaker corrections Download instructions