Telefónica submits ASR system for Speech-COSER Challenge 2026
Original title: TID System for Speech-COSER Challenge 2026: ASR Track
An ensemble of seven LoRA systems reached 22.17% normalized word error rate on elderly rural Spanish speech.
Telefónica Innovación Digital describes its automatic speech recognition entry to the Speech-COSER Challenge 2026. The track uses spontaneous speech from elderly speakers recorded in rural Spain. CTC-based pseudo-forced alignment scores audio-text match quality to guide a two-stage LoRA curriculum, followed by magnitude pruning of the LoRA matrices. ROVER combines seven systems.
Unconfirmed: Word error rates are the authors' reported challenge results.
- Authors report 36.25% unnormalized and 22.17% normalized WER.
- Post-hoc magnitude pruning of LoRA matrices is used to improve generalization.
- Alignment scores guide a two-stage LoRA curriculum.
- Organizations
- Telefónica Innovación Digital
- Countries
- Spain
From the original article
1 min read at arxiv.orgWe describe the system submitted by Telefónica Innovación Digital to the Speech-COSER Challenge 2026, which features spontaneous speech from elderly speakers recorded in rural Spain. For ASR, we perform CTC-based pseudo-forced alignment scores audio-text match quality to guide a two-stage LoRA curriculum, followed by post-hoc magnitude pruning of the LoRA matrices to enhance generalization.
By Fernando López · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.