Basque speech language model enables fast spoken interaction
Developing a Speech Language Model for Basque
Computation and LanguageArtificial Intelligence
Summary
Basque is a minority language that often lacks advanced technology like smart voice assistants. This work creates the first open speech language model specifically for Basque that can understand and respond quickly in spoken conversations. To deal with limited data, the authors translated a large English speech dataset into Basque and generated synthetic Basque audio. Tests show the system produces understandable, high-quality speech, proving that such tools can be built even for less common languages using smart techniques.
What this means in practice
- •For voice assistant developers: Build Basque-speaking voice assistants with low-latency, end-to-end speech understanding and generation.$Commercial implications: Enables creation of Basque voice assistant products that previously lacked robust spoken interaction capabilities.
- •For language technology providers: Develop speech-based tools for minority languages by fine-tuning modular models using synthetic translated datasets.
Authors
Andoni Sudupe, Inma Hernaez, Ander Arriandiaga, Christoforos Souganidis, Ibon Saratxaga, Eva Navas
Abstract
Although the recent development of LLMs has paved the way for multi-modal interaction, minority languages face difficulties keeping pace. This work presents the first open Speech Language Model for Basque, designed for low-latency, end-to-end spoken interaction. Based on the LLaMA-Omni architecture, the system integrates a Latxa model as its backbone, transferring its Basque capabilities to the spoken modality by fine-tuning several modules. To address data scarcity, the English VoiceAssistant-400K corpus was translated into Basque, and the audio was synthesized using a Basque TTS. Evaluation results confirm the system's viability, achieving 3.86/5 in content quality, a 9.94% WER for spoken intelligibility, and a 3.07 UTMOS for audio quality. This work demonstrates that advanced spoken interaction systems can be developed using modular architectures and synthetic data even in low-resource languages.