Aprimoramentos de performance de modelos LLM para aplicações de transcrição de texto em sistemas elétricos de potência

Authors

  • Victor Hideki Yoshizumi
  • Sofia Moreira De Andrade Lopes
  • Danilo Hernane Spatti
  • Rogério Andrade Flauzino
  • Ivan Nunes da Silva
  • Ivan Gídaro Ricci
  • Alexandre Latorre
  • Celso Moreira de Lima Junior

DOI:

https://doi.org/10.29327/1863744.1-29

Abstract

The use of speech transcribers in everyday operations has grown in recent periods thanks to the increasingly efficiency in models. Such transcribers can also be combined with Generative Artificial Intelligence models, such as chat GPT, further increasing their performance. However, in certain cases, given the particularity of the words and sentences to be transcribed, improvements with personalized vocabularies are necessary, allowing specific instructions to be directed to the chat GPT and thus improving transcription results. This process is known as Prompt Engineering and was used in this work to considerably improve the transcription of telephone conversations between the ONS and the transmission operator.

Downloads

Published

2024-10-18

Issue

Section

Articles