Cross-lingual Embedding Clustering for Hierarchical Softmax in Low-Resource Multilingual Speech Recognition
arXiv:2501.17615v2 Announce Type: replace Abstract: We present a novel approach centered on the decoding stage of Automatic Speech Recognition (ASR) that enhances multilingual performance, especially for low-resource languages. It utilizes a cross-lingual embedding clustering method to construct a hierarchical Softmax (H-Softmax) decoder, which enables similar tokens across different languages to share similar decoder representations. It addresses the limitations of the previous Huffman-based H-Softmax method, which relied on shallow features in token similarity assessments. Through experiment
The continuous drive for more inclusive and efficient AI models necessitates overcoming language barriers, especially for less resourced languages, in fields like speech recognition.
This development significantly lowers the barrier for leveraging AI in diverse linguistic contexts, expanding AI's global utility and reducing digital divides across languages.
Multilingual speech recognition systems can now be significantly more accurate and resource-efficient for low-resource languages, fostering greater accessibility and broader deployment.
- · Low-resource language communities
- · Multilingual AI developers
- · Global technology platforms
- · AI research institutions
- · Monolingual AI solutions
- · Systems heavily reliant on resource-intensive, language-specific models
Improved performance and reduced computational overhead for multilingual Automatic Speech Recognition (ASR) systems.
Accelerated development and adoption of voice interfaces and AI services in previously underserved linguistic markets.
Enhanced data collection and transcription for hundreds of languages, accelerating the development of more generalized and culturally relevant AI.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL