Xiaomi Group's AI Lab released the ZipVoice series of speech synthesis models.

On September 12th, according to Xiaomi technology news, recently, the new generation Kaldi team of Xiaomi Group's AI Lab released the ZipVoice series speech synthesis model based on the Flow Matching architecture - ZipVoice and ZipVoice-Dialog. ZipVoice solves the pain points of existing zero-sample speech synthesis models with large parameter amounts and slow synthesis speeds, while ZipVoice-Dialog addresses the bottlenecks in stability and inference speed of existing dialogue speech synthesis models.