Alexey Korepanov

`torch.arange` doesn't implemented for `float16`

7

# Problem 👩‍⚕️ I found that when `x` is in `float16` this line doesn't work: https://github.com/speechbrain/speechbrain/blob/c7219b5b2b5c1b8a68da23e81eae0046b6769556/speechbrain/nnet/attention.py#L345 It is important because when I train the model in `fp16` (with `autocast` for...

Hi! I noticed strange behavior of `DynamicItemDataset`. # Problem 👩‍⚕️ My dataset looks like this: ``` { "my_audio.wav": { "audio_path": "my_audio.wav", "ref": "example of speech", "length": 5 } } ```...

Why `.float()` is in `conformer_modules.py`?

2

https://github.com/NVIDIA/NeMo/blob/eae1684f7f33c2a18de9ecfa42ec7db93d39e631/nemo/collections/asr/parts/submodules/conformer_modules.py#L259 Hi! I'm wondering why `.float()` is here. It seems like it will break code if both model parameters and input in float16

How to make `irfft` in ONNX

12

# Ask a Question ### Question Hello! Can you point out how `irfft` can be used? I found issues and documentation on using `rfft`, but didn't find anything about `irfft`....

I've made a torch reimplementation for both offline and streaming implementation. Would you be interested in accepting this contribution?

47

Hello! I wrote a complete torch implementation without using rust or tract. I'm thinking about PR. So I decided to ask you, would you be interested in accepting this contribution?...

Raw examples of running DeepFilterNet3 on the Web (in JS)

6

Hello! Recently we've merged PR https://github.com/Rikorose/DeepFilterNet/pull/452 with WASM conversion support. I've now uploaded a couple of raw examples using DeepFilterNet3 on the web. One of them using `wasm` and the...

Pure torch implementation

26

I've created this issue about pure torch reimplementation - https://github.com/Rikorose/DeepFilterNet/issues/430 Sharing code. This is draft PR, so right now work still in progress, and i can make some changes later....

RNNoise v0.2 does not compile on Mac and Linux

8

I'm doing similar steps that worked for me: ``` git clone https://github.com/xiph/rnnoise.git cd rnnoise git checkout 904a876dce1f9ab8860c0a5000ed151f9f6eef58 ./autogen.sh ./configure make ``` `On Linux` on make step I recieve a lot...

`torchaudio.info.num_frames` can give wrong results so it can provide false exceptions

4

### Tested versions current master ### System information linux ### Issue description I get here wrong `num_frames` on m4a opus-files https://github.com/pyannote/pyannote-audio/blob/cd3f550d00ea6bfb155dc7aef17e4f9c2516ee55/pyannote/audio/core/io.py#L280 Repro and described error can be founded here: https://github.com/pytorch/audio/issues/3731...

assign_word_speakers fix

1

Hello! Seems like here is mistake. We should look for closest instead of sum. Simple counterexample: ``` predicted_sample - [0,1] diarization_segments: SPEAKER_00: [2, 3], [4, 5] SPEAKER_01: [3, 4] ```...