Software for Automatic Speech Recognition via Whisper models applied to Oral History interviews in the Portuguese language

2026-04-12 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, quick

Summary

Ethos AT is a new desktop software designed for automatic speech transcription, utilizing OpenAI Whisper models for local processing to ensure data privacy and accessibility. Developed for non-programming experts, such as oral history researchers, it facilitates transcription of Portuguese language interviews. A comparative analysis was performed on six Whisper models: small, medium, large, large-v2, large-v3, and turbo. The study found that larger models offered higher lexical accuracy, while smaller models provided faster execution with acceptable quality. The turbo model achieved an effective balance between accuracy and speed, making Ethos AT a secure, efficient, and user-friendly solution for academic and institutional contexts.

Key takeaway

For oral history researchers or academic institutions needing secure, efficient transcription of Portuguese audio, Ethos AT offers a user-friendly desktop solution. You should consider the Whisper turbo model for its effective balance of accuracy and speed, or larger models like large-v3 if maximum lexical accuracy is paramount, despite longer processing times. This tool ensures data privacy through local processing, a critical factor for sensitive research.

Key insights

Ethos AT provides private, accessible automatic speech transcription for Portuguese using Whisper models.

Principles

• Larger Whisper models yield higher accuracy.
• Smaller Whisper models offer faster transcription.
• Local processing enhances data privacy.

Method

Six Whisper models (small, medium, large, large-v2, large-v3, turbo) were compared for transcription accuracy, error types, and processing time on Portuguese oral history interviews.

In practice

• Use Whisper turbo for balanced accuracy/speed.
• Prioritize larger models for maximum accuracy.
• Choose smaller models for faster processing.

Topics

• Ethos AT
• Automatic Speech Recognition
• OpenAI Whisper Models
• Oral History
• Portuguese Language Processing

Best for: AI Engineer, NLP Engineer, Research Scientist, Domain Expert, AI Scientist, Machine Learning Engineer

Related on AIssential

• BaltiVoice: A Speech Corpus and Fine-tuned Whisper ASR System for the Balti Language — Fine-tuning pre-trained ASR models with modest, newly created corpora can establish functional speech recognition for low-resource languages.
• Local Whisper Audio Transcription — Faster-Whisper enables fast, private, and local audio transcription using optimized Whisper models in Python.
• Certas Palavras: A 1980s-90s Brazilian Radio Corpus to Test TTS Models in Noisy Multi-Speaker Dialogues — Certas Palavras offers a challenging Brazilian Portuguese dataset for robust TTS training in noisy, spontaneous environments.
• Semi-Supervised Speech Confidence Detection using Pseudo-Labelling and Whisper Embeddings — A novel framework detects speaker confidence by fusing human-engineered features and Whisper embeddings, enhanced by pseudo-labelling for data scarcity.
• Speech intelligence for enterprises - Mistral AI — Mistral AI offers open, customizable speech intelligence models for enterprises, enabling human-like voice interactions and accurate transcription.
• Detecting Stuttering with Artificial Intelligence: A Hybrid Method for Brazilian Portuguese — A hybrid AI system automates stuttering detection and classification in Brazilian Portuguese with high accuracy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.