SURT

The article presents an improved version of SURT (Streaming Unmixing and Recognition Transducer) for speaker-attributed transcription in multi-talker speech recognition. The authors propose methods for both short mixtures and long…