This study proposes a lightweight method to detect code-switching directly from speech by identifying language-selective hidden-state dimensions in pretrained encoders and aggregating frame-level activations. Achieving a mean F1 of 0.798 on CS-FLEURS with Whisper-small, it outperforms CNN and HuBERT baselines, offering an efficient tool for building representative multilingual corpora without relying on transcription.
Code-switching
代码切换
The phenomenon of naturally mixing multiple languages within a single sentence
Curated Intelligence
The Technology Innovation Institute (TII) has released Falcon ASR, an open-source speech recognition model designed to address the low-resource challenges of Arabic ASR. Built upon the Falcon LLM architecture, it is optimized for Arabic dialects and code-switching scenarios, providing high-precision, customizable local voice interaction infrastructure for the Middle East.
Zhipu AI has open-sourced GLM-ASR-Nano, a 1.5B parameter speech recognition model designed for robustness in real-world scenarios. It achieves an average error rate of 4.10, significantly outperforming Whisper V3 (6.93) on open-source benchmarks, with exceptional performance in dialects, low-volume speech, and code-switching. The model utilizes an asymmetric 'heavy perception, extreme compression' architecture and is already deployed in Zhipu's production input method.