Banco Santander deploys scalable Speech-to-Text system to automate call center transcription

Banco Santander, one of the world's leading banks, faced the challenge of managing a large volume of customer service calls. The bank needed a Speech-to-Text (STT) system capable of accurately diarizing audio transcriptions to improve service quality and reduce response times across its call centers.

Dive conducted a benchmark of 17 STT technologies under various business and technical criteria, selecting Azure (public cloud), Speechmatics (service), and OpenAI Whisper medium (open-source) as top options. A scalable, customizable STT system was designed and validated through a proof of concept using real call center data, employing deep learning for speech recognition and Generative AI to augment training data volumes.

The implemented STT system achieved accurate transcription of customer calls, allowing agents to redirect effort toward higher-value tasks. The system's scalability enabled Banco Santander to handle growing call volumes seamlessly. The bank specifically noted improvements in transcription accuracy and the solution's ability to process large volumes of audio data.

• Benchmarking a broad set of STT technologies (17 options) before selection ensures the chosen solution meets both business and technical requirements at scale.
• A hybrid approach combining cloud, SaaS, and open-source STT options provides deployment flexibility across different productive environments.
• Generative AI can supplement real training data, improving model performance for specialized financial call center transcription tasks.