When doing automatic speech recognition (ASR) on audio/video, the transcript often lacks proper punctuation. This is especially true when using free ASR services like Google Cloud Speech-to-Text. Jennifer Marie demonstrates how to use ChatGPT to add proper punctuation to a transcript and improve it’s quality in general.
Diagram of the process:
graph LR
A[Audio] -- Virtual Audio Cable --> C[Google Docs]
DD[Specific Prompt] --> E[ChatGPT]
C -- Copy --> E
F{All text transcribed?}
E --> F -->|Yes| G[Google Docs >\n Compare documents]
F -->|No| E
The key steps of the process are:
- Take the raw transcript, for example, prepared with Virtual Audio Cable and Google Docs
- Copy it to ChatGPT with a specific prompt
- If ChatGPT stops midway, find your spot in the original, copy paste into new ChatGPT prompt
- Paste ChatGPT outputs into a new Google Doc to combine the full properly punctuated transcript
- Use Google Docs compare documents feature to validate ChatGPT corrections vs original
As Virtual Audio Cable Jeniffer refers to VB-Audio software.
Specific Prompt:
You are a professional transcriptionist. You have perfect spelling, punctuation, and know how to format transcripts correctly. I will paste a text for you to correct. You will add appropriate capital letters, periods, commas, apostrophes, and question marks where necessary, and other punctuation marks needed. You will NOT make any other edits to the text. You will not change any words. You will keep typing until the entire text is punctuated. Here is the text: ""