Release v1.4.0

1. 非流式识别： 复现silero-vad 并适配gpu @lovemefan
2. 感谢 @Ivy233 贡献的流式识别

Release V1.3.1

1. fix IM2COL operator bug in CUDA and Metal, faster about 30% on gpu than mixed computation
2. bug fix : avoid j out of index when compute fbank @lovemefan

Tanks ggml-org and @Ivy233 ggml-org/ggml#991

Release V1.3.0

🎉🎉🎉Support more GPU backend (hybrid cpu and gpu backend because of im2col operator), such as CUDA, Vulkan, CANN (not tested without ascend machine)

v1.2.0

1. Support Matel backend （hybrid cpu and gpu backend）
2. fix bug about CUDA norm by @Ivy233 #18
3. fix tesnor set error by @pockers21 #11

v1.1.0

Support Q3, Q4, Q5, Q6, Q8 quantization model.