Prabhsimran Singh

Implement [PyTorch Transformer LM lattice rescoring](https://danielpovey.com/files/2021_icassp_parallel_lattice_rescoring.pdf) in the Decoder class of kaldi-serve [core C++ library](https://github.com/skit-ai/kaldi-serve/tree/master/src). Please find a similar implementation for RNN LM lattice rescoring [here](https://github.com/skit-ai/kaldi-serve/blob/master/src/decoder/decoder-common.cpp#L21). The idea here is...

Hi @al-zatv not yet, but we will be adding support (more granular control) soon, will link relevant version here. As for `frame_subsampling_factor`, it is [3 by default](https://github.com/Vernacular-ai/kaldi-serve/blob/master/src/config.hpp#L32) but you can...

### Task 1 Implement [Speech Endpointing feature](https://kaldi-asr.org/doc/structkaldi_1_1OnlineEndpointRule.html#:~:text=By%20endpointing%20in%20this%20context,decide%20when%20to%20stop%20decoding.) in the [Decoder class](https://github.com/skit-ai/kaldi-serve/blob/master/src/decoder/decoder.cpp) of kaldi-serve [core C++ library](https://github.com/skit-ai/kaldi-serve/tree/master/src). Speech Endpointing is the process of deciding when to stop decoding based on trailing...

Improve on the current confidence scoring algorithm [here](https://github.com/skit-ai/kaldi-serve/blob/master/include/kaldiserve/decoder.hpp#L190). The current confidence scores aren't very informative in terms of relative importance of the resulting alternatives. This is because Kaldi's AM and...

For python2, you will need to do this instead: ``` import imp data = imp.load_source('data.twitter.data', 'data/twitter/data.py') ```

@cavin12400 It should be running on GPU by default. Please check your `tensorflow-gpu` installation.

@nidhikamath91 sorry I'm not very familiar with TensorFlow serving. You'd be better off posting this on their issues itself. Although this seems like CLI argument error.

@nidhikamath91 according to what I could gather from the MNIST example on TF serving, I think your `tensor_info_y` needs to be the score outputs or in this case the softmax...

@nidhikamath91 your def of x looks fine to me. However I'm not sure this is gonna work since we're feeding the encoder `state` to the decoder during inference then feeding...

@nidhikamath91 we're manually converting the input query into `token ids` that are fed into the encoder as `encode_seqs2` and then feed the encoder `state` to the decoder to decode in...