Shanghai Jiao Tong University Partners with ByteDance to Launch LSLM: AI Voice Model Achieving Breakthrough in 'Listen and Speak'

The LANCE Laboratory at Shanghai Jiao Tong University has collaborated with ByteDance to develop an innovative voice interaction model called LSLM, also known as 'Little L'. This model excels in real-time interaction, noise resistance, and the ability to recognize new speakers, approaching the naturalness of human conversation. LSLM employs an end-to-end design, encompassing auditory and vocal channels, utilizing decoder-only TTS to generate speech, and incorporating streaming self-supervised learning to process audio input in real time. Its unique features include full-duplex modeling, enabling interruptions and turn-taking during conversations; strong noise resistance that maintains stability.