zycv / openspeaker Goto Github PK

Hi,
Which model is used for embedding extraction ? Is it ECAPA-TDNN ?

Regards

Almost most of different audio files get cosine similarity over 0.9 ?
How to interpret the similarity value?

In file included from /home/stuart/OpenSpeaker/build/_deps/googletest-src/googletest/src/gtest-all.cc:42:
/home/stuart/OpenSpeaker/build/_deps/googletest-src/googletest/src/gtest-death-test.cc: In function ‘bool testing::internal::StackGrowsDown()’:
/home/stuart/OpenSpeaker/build/_deps/googletest-src/googletest/src/gtest-death-test.cc:1301:24: error: ‘dummy’ may be used uninitialized [-Werror=maybe-uninitialized]
 1301 |   StackLowerThanAddress(&dummy, &result);
      |   ~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~

appended -Wno-maybe-uninitialized then compiles

CMake Error at /usr/local/lib/cmake/gflags/gflags-nonamespace-targets.cmake:37 (message):
Some (but not all) targets in this export set were already defined.

Targets Defined: gflags_nothreads_static

Targets not yet defined: gflags_static

speechbrain训练好的模型使用convert_to_torchscript.py转换时出现以下错误error，求问如何修改？

torch.jit.frontend.UnsupportedNodeError: try blocks aren't supported:
File "/home/wieneralan/projects/speechbrain/speechbrain/lobes/models/Xvector.py", line 101

    for layer in self.blocks:
        try:
        ~~~ <--- HERE
            x = layer(x, lengths=lens)
        except TypeError:

wechat: znsoft 验证信息：openspeaker

so this mearningful work will be noticed by more guys, ths.

Dear author:
I try to compare your fbank and torchaudio fbank. For an input with shape [1, 16000], the output of yours is [1, 98, 80], however, the torchaudio get [101, 80]. I guess some of padding is different between yours and torchaudio. Could you give some tips on how to align these two implementations? thank you very much.

OpenSpeaker/voiceprint/voiceprint_main.cc
line 89: OpenSpeaker::Features features(sample_rate, feats_dims);

头文件里的定义是
namespace openspeaker {

大小写不匹配，编译的时候报错了，改成统一的就可以了。