Bringing NURC/SP to Digital Life: the Role of Open-source Automatic Speech Recognition Models

TL;DR

This paper evaluates and compares several automatic speech recognition models trained on spontaneous and prepared Portuguese speech to transcribe a large Brazilian linguistic corpus, improving transcription accuracy for spontaneous speech analysis.

Contribution

It provides an evaluation and error analysis of multiple ASR models on spontaneous Portuguese speech, selecting the best model for large-scale automatic transcription of the NURC/SP corpus.

Findings

Best model achieved lower WER and CER scores.

Automatic transcription covered 284 hours of speech.

Error analysis identified key challenges in spontaneous speech recognition.

Abstract

The NURC Project that started in 1969 to study the cultured linguistic urban norm spoken in five Brazilian capitals, was responsible for compiling a large corpus for each capital. The digitized NURC/SP comprises 375 inquiries in 334 hours of recordings taken in S\~ao Paulo capital. Although 47 inquiries have transcripts, there was no alignment between the audio-transcription, and 328 inquiries were not transcribed. This article presents an evaluation and error analysis of three automatic speech recognition models trained with spontaneous speech in Portuguese and one model trained with prepared speech. The evaluation allowed us to choose the best model, using WER and CER metrics, in a manually aligned sample of NURC/SP, to automatically transcribe 284 hours.