-
Hi,
I'm currently trying to train a transducer model using rf. I use the `torch_amp="bfloat16"` option from previous setups. In the predictor I use a `rf.LayerNorm` followed by `rf.LSTM`. I think thi…
-
这要训练多少轮效果才好,感觉损失都好高,效果很差
-
I'd like to have a way to penalize symbol delay in RNN-T computations, and to measure it during training.
(This is relevant for systems that use time masking, to avoid the time-masking encouraging th…
-
**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_…
-
Does the sherpa server support VAD or endpointing?
-
Hi guys,
I am trying to replicate the results from the gigaspeech recipe, as a comparison to other models we trained before.
The pretrained model was obtained from [Gigaspeech repo](https://hug…
-
## 🐛 Bug
I made my own wav2vec2.0 model, and pretrained it with some data.
model training was successful, and inference is working with the command written in wav2vec2.0 README.md.
But the problem …
-
## 🐛 Bug
When inference using wav2vec_seq2seq, the decoder is not compatible with the model
### To Reproduce
Steps to reproduce the behavior (**always include the command you ran**):
I ru…
-
thanks very much for your great project!
I have two questions to ask:
1. how big is the the transducer loss for a well performed model? or the model is converged?
2. is there an…
-
I am getting a lot of deletions in my RNN-T training/decoding setup relative to Transformer/Conformer. The data is the "Malach" corpus; about 200 hours of English but accented speech from Holocaust su…