-
The formular for gradient is below in` warprnnt_numba` and `warp_transducer cpu`:
```
T, U, _ = log_probs.shape
grads = np.full(log_probs.shape, -float("inf"))
log_like = betas[0, 0] …
-
## Bug Description
I get an error when converting a conformer transducer enecoder to tensorrt. (asr task)
## To Reproduce
[requirenments.txt](https://github.com/pytorch/TensorRT/files/123430…
-
Hi @titu1994,
Following our discussion in [this thread](https://github.com/NVIDIA/NeMo/discussions/7010#discussioncomment-9736124), I’m training a cache-aware FastConformer hybrid CTC-RNNT model fo…
-
When just setting up the RNNT benchmark using the run.sh with the stage set to -1. It fails with python import error of the mlperf_loadgen.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZNSt15__e…
-
## 🐛 Bug
Getting an assertion error (cutoff) when trying to run Wav2Vec 2.0 CTC inference with Transformer LM. What do I need to change to get this working?
### To Reproduce
Steps to reproduc…
-
Hi,
I am using conformer ctc with HLG decoding for streaming. I have used the implementation mentioned in #1218 online_decode.py. While decoding longer calls I am getting higher latency. In this…
-
Transducer class 코드에서.
Q1.
pred != 0 ---> 이 의미는 blank가 아닌 character를 예측한 것으로 transducer 구조(첨부한 그림 참조)에서 수직으로 이동한 것에 해당합니다.
이 경우에는 (encoder time step) t가 증가하면 안되는 것 아닌가요?
![image](https://user-ima…
-
**Describe the bug**
When I tried to generate timestamp with model: nvidia/parakeet-tdt-1.1b, I got following error,
ValueError: `char_offsets`: [{'char': [tensor(607, dtype=torch.int32)], 'start_…
-
**Describe the bug**
我们使用wenet中实现的bestrq代码,试图复现预训练模型的结果,但是效果不好,完全达不到论文中的WER结果。
**To Reproduce**
1. 我们使用librispeech 960h数据进行预训练模型训练,使用10h和100h的数据进行微调。
2. 我们仿照wav2vec2的实验设置,预训练模型接CTC,而不是bestrq论文中的…
-
I have a Japanese model using the streaming Zipformer trained with the command below, and I noticed a (~1.0%) difference between my decoding results with decode.py and streaming_decode.py.
> ./prun…