Guide for Using Prompt Words in MemoAI

Transcribing speech to text with Memo is not just about inputting a piece of audio and transcribing it into text. Considering that the Memo model is based on Whisper, the corresponding Whisper also provides the ability of prompt words, which is generally at the level of GPT-2.

Scope of Effect

Prompt words are generally only effective for Medium and Large models. The effectiveness for models below these is unpredictable.

Common Scenarios

You can choose either Chinese or English below, but English will generally yield better results.

Adding Punctuation

"This is a lecture on xx, please include punctuation when transcribing."

Importing Professional Vocabulary

You can import professional vocabulary through prompt words for accurate recognition by the AI model.

"This is a lecture on xx, it contains the following terms: 'computer' and 'shrimp dumpling'."

Transcribing Songs with Background Music

"Please ignore the background music and transcribe the lyrics."

Precise Sentence Segmentation

"Please add a period."

Conclusion and Related Materials

In summary, whenever facing any issues, prioritize using prompt words. These help guide the AI model on how to process audio.

• Discussion on Prompt Words
• Whisper Prompting Guide