Powered by OpenAIRE graph
Found an issue? Give us feedback
addClaim

Da voz aos sentimentos: Transcrição automática com vista ao controlo de qualidade de chamadas telefónicas

Authors: Morais, Inês Filipa Correia;

Da voz aos sentimentos: Transcrição automática com vista ao controlo de qualidade de chamadas telefónicas

Abstract

Este estudo explora o uso de tecnologias de transcrição automática de fala e análise de sentimento no controlo de qualidade de chamadas de um contact center de uma seguradora. Foram comparadas duas ferramentas de transcrição – WhisperX e Microsoft Azure Speech-to-Text – com o objetivo de avaliar a viabilidade de soluções de código aberto para substituir alternativas comerciais dispendiosas. Os resultados mostram que, apesar de ambos os modelos gerarem transcrições compreensíveis, o WhisperX destaca-se pela fluidez e precisão em captar nuances linguísticas e variações regionais do português. No entanto, enfrenta desafios em contextos de áudio complexos, como múltiplos locutores e variações de intensidade. Além disso, foram propostas diversas aplicações práticas das transcrições para otimizar o atendimento ao cliente, incluindo a análise de cumprimento de guiões, métricas de diálogo e deteção de sentimentos. Para demonstrar o potencial destas aplicações, foi implementado o modelo FinBERT-PT-BR nas transcrições, que permitiu captar sentimentos predominantes nas interações e identificar momentos de insatisfação e incerteza. Os principais contributos deste trabalho incluem a análise de desempenho de ferramentas de transcrição aplicadas a interações em português europeu, bem como a proposta de aplicação de um modelo de análise de sentimento orientado para a melhoria da qualidade do atendimento ao cliente. Entre as limitações, destaca-se a necessidade de dados anotados para uma avaliação mais rigorosa. Futuras investigações poderão beneficiar de uma abordagem multimodal que combine transcrição e características acústicas para uma análise emocional robusta, especialmente num contexto de contact center.

This study explores the use of automatic speech transcription and sentiment analysis technologies in the quality control of call monitoring in an insurance contact center. Two transcription tools – WhisperX and Microsoft Azure Speech-to-Text – were compared to evaluate the feasibility of open-source solutions as alternatives to costly commercial options. The results show that, while both models generate comprehensible transcriptions, WhisperX stands out for its fluency and accuracy in capturing linguistic nuances and regional variations of Portuguese. However, it faces challenges in complex audio contexts, such as multiple speakers and varying intensity levels. Furthermore, several practical applications of transcriptions were proposed to optimize customer service, including script compliance analysis, dialogue metrics, and sentiment detection. To demonstrate the potential of these applications, the FinBERT-PT-BR model was implemented on the transcriptions, which enabled the capture of predominant sentiments in interactions and identification of moments of dissatisfaction and uncertainty. The main contributions of this work include a performance analysis of transcription tools applied to european portuguese interactions, as well as the proposal of a sentiment analysis model aimed at improving customer service quality. Among the limitations, the need for annotated data for more rigorous evaluation can be highlighted. Future research may benefit from a multimodal approach that combines transcription and acoustic features for a more robust emotional analysis, particularly in a contact center context.

Country
Portugal
Keywords

Chamadas telefónicas, Reconhecimento automático da fala, Automatic speech recognition, WhisperX, Automatic transcription, Domínio/Área Científica::Engenharia e Tecnologia::Outras Engenharias e Tecnologias, Análise de sentimento, Contact center, Sentiment analysis, Domínio/Área Científica::Ciências Sociais::Economia e Gestão, Transcrição automática, Phone calls

  • BIP!
    Impact byBIP!
    selected citations
    These citations are derived from selected sources.
    This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    0
    popularity
    This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
    Average
    influence
    This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    Average
    impulse
    This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
    Average
Powered by OpenAIRE graph
Found an issue? Give us feedback
selected citations
These citations are derived from selected sources.
This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Citations provided by BIP!
popularity
This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
BIP!Popularity provided by BIP!
influence
This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Influence provided by BIP!
impulse
This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
BIP!Impulse provided by BIP!
0
Average
Average
Average
Upload OA version
Are you the author of this publication? Upload your Open Access version to Zenodo!
It’s fast and easy, just two clicks!