Text-to-speech modeling

Spark-TTS: Redefining the Balance between Efficiency and Sound Quality in Speech Synthesis Spark-TTS is an innovative text-to-speech (TTS) model developed by the SparkAudio team. Its core is based on the BiCodec architecture and large-scale language model (LLM) technology, which realizes a breakthrough in efficiency and sound quality in the field of speech synthesis. First, the technical architecture: single-stream decoupled speech coding BiCodec design principle Spark-TTS through the proposed BiCodec encoder, the speech signal is decomposed into two types of complementary tokens: low-bit-rate semantic tokens: focusing on ...