how to keyterm prompt nova 3
# ask-ai
s
how to keyterm prompt nova 3
t
Thanks for reaching out, A. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.
Keyterm prompting is available for English transcription using the Nova-3 model. To use it: • Utilize the
keyterms
parameter, which is a list of key terms to improve recognition accuracy. • Ensure you are using a Nova-3 model (models that start with "nova-3"). • Set the language to English ("en" or "en-US"), as keyterm prompting only works with English for Nova-3 models. If you are using other models (Nova-2, Nova-1, Enhanced, or Base), you should use the Keywords feature instead of keyterms. Sources: GitHub::livekit::agents | livekit-plugins/livekit-plugins-deepgram/livekit/...
s
and for dutch? i cant use keyterms parameter? @livekit assisant
t
Hey there! I totally understand the confusion. Just to clarify, Nova-3 models can only use keyterm prompting for English transcriptions right now. If you're working with Dutch transcription, you'll need to explore some other options. I'm here to help you figure out the best solution, so don't hesitate to reach out if you have any questions! Sources: GitHub::livekit::agents | livekit-plugins/livekit-plugins-deepgram/livekit/...
s
what alternate options do i have?
t
Hey there! I totally understand your need for Dutch transcription. While Nova-3 models won't let you use keyterm prompting, don't worry - we've got some great alternatives. You could try switching to Nova-2, Nova-1, Enhanced, or Base models, which all support the Keywords feature. This will help you highlight and recognize those specific words you're looking for. Would you like me to walk you through how to do that? I'm here to help make this as smooth as possible for you! Sources: GitHub::livekit::agents | livekit-plugins/livekit-plugins-deepgram/livekit/...
s
how to add additional keywords to nova-2 in the code?
t
To add extra keywords to Nova-2 models, I've got a super handy trick for you! 🌟 When you're setting up your speech-to-text with the STT class, you can use the keywords parameter to give certain words some special attention.
Copy code
from livekit.plugins.deepgram import STT

deepgram_stt = STT(
    model="nova-2-general",
    keywords=[("LiveKit", 1.5), ("YourKeyword", 2.0)],
    # other parameters...
)
Here's the scoop on how this works: • Each keyword is like a little booster pack in a tuple - you get the word and its boost value. • Boost those keywords up with positive values to help them shine, or dial them down with negative values if you want them to take a step back. • If you don't specify anything, it defaults to an empty list
[]
- totally optional! Need any more help understanding this? I'm right here and happy to walk you through it! Just give me a shout. 😊 Sources: livekit.plugins.deepgram API documentation | Deepgram integration guide | LiveKit Docs