FFSTC: Fongbe to French Speech Translation Corpus

TL;DR

This work introduces FFSTC, the first public end-to-end Fongbe–French speech translation corpus, totaling about $31$ hours of spoken Fongbe with French text. The dataset is created through a multi-source collection and a rigorous, multi-tier validation process, resulting in a high-quality resource with $16{,}447$ sentences and diverse vocabulary. Baseline end-to-end translations are established using Fairseq transformer_s and conformer models, achieving BLEU scores of $8.96$ and $8.14$, respectively, providing a benchmark for future improvements. The paper demonstrates the feasibility and value of targeted, end-to-end SL datasets for under-resourced, tonal languages and outlines clear directions for leveraging pre-training and transfer learning to advance performance in West African language translation scenarios.

Abstract

In this paper, we introduce the Fongbe to French Speech Translation Corpus (FFSTC) for the first time. This corpus encompasses approximately 31 hours of collected Fongbe language content, featuring both French transcriptions and corresponding Fongbe voice recordings. FFSTC represents a comprehensive dataset compiled through various collection methods and the efforts of dedicated individuals. Furthermore, we conduct baseline experiments using Fairseq's transformer_s and conformer models to evaluate data quality and validity. Our results indicate a score of 8.96 for the transformer_s model and 8.14 for the conformer model, establishing a baseline for the FFSTC corpus.

Figures (1)

• Figure 1: Vocabulary and number of words of each split