Binbin Zhang
Binbin Zhang
There is no such problem in our experiment. @Mddct @xingchensong @day9011 Any idea on the problem?
Any difference of the test data on v1/v2 test? Is there any long audio in the test?
Just use multiple utterances with different lengths to warmup, it should work.
What kind of checking? It's better to upgrade to 1.12 since it is the lastest stable version.
No, it will take too mush efforts to check, just go.
评估一下对不含热词的测试集的影响
那结果看起来不错，这里你把 1.0 改成一个参数，放到这个位置 https://github.com/wenet-e2e/wenet/blob/main/runtime/core/decoder/context_graph.h#L34。 ，默认是 0.0，这样不影响以前的功能。 可以命名为 `incremental_context_score`.
any update?
Great job!
How about it now? We do not have the arm linux platform. If you make it works, could you help to contribute arm linux support in WeNet?