search

tensorflow / models

Models and examples built with TensorFlow
Other
77.24k stars 45.75k forks source link

DeepSpeech model accuracy #7567

Open BuaaAlban opened 5 years ago

BuaaAlban commented 5 years ago

The speech trained by the code released here cant' reach high accuracy, WER reach to about 0.20, and I found a problem that it can't translate words like "moss-->mos mirror-->miror", any repeated character can't be translated right.

tensorflowbutler commented 5 years ago

Thank you for your post. We noticed you have not filled out the following field in the issue template. Could you update them if they are relevant in your case, or leave them as N/A? Thanks. What is the top-level directory of the model you are using Have I written custom code OS Platform and Distribution TensorFlow installed from TensorFlow version Bazel version CUDA/cuDNN version GPU model and memory Exact command to reproduce

BuaaAlban commented 5 years ago

Thank you for your post. We noticed you have not filled out the following field in the issue template. Could you update them if they are relevant in your case, or leave them as N/A? Thanks. What is the top-level directory of the model you are using tensorflow/models/deep_speech

Have I written custom code Yes

OS Platform and Distribution Ubuntu 16 TensorFlow installed from pip TensorFlow version 1.14 Bazel version ??? CUDA/cuDNN version 10.0 /7.5.0 GPU model and memory

Exact command to reproduce

YuanEric88 commented 4 years ago

@BuaaAlban

Hi, I am also training deepspeech by libirspeech dataset, and I have some questions as follow: 1, What is the training dataset in your experiment? Is it the combination of train-clean-100, train-clean-360 and train-other-500? or just part of them? 2, In which dataset did you achieve WER=0.2, dev-clean only, dev-clean + dev-other, or test-clean? 3, what is the training strategy of your experiment? Specifically, the value of learning_rate, training_epochs, learning_rate_decay strategy, etc.

Looking forward to your reply. Thanks.

BuaaAlban commented 4 years ago

@BuaaAlban

Hi, I am also training deepspeech by libirspeech dataset, and I have some questions as follow: 1, What is the training dataset in your experiment? Is it the combination of train-clean-100, train-clean-360 and train-other-500? or just part of them? 2, In which dataset did you achieve WER=0.2, dev-clean only, dev-clean + dev-other, or test-clean? 3, what is the training strategy of your experiment? Specifically, the value of learning_rate, training_epochs, learning_rate_decay strategy, etc.

Looking forward to your reply. Thanks.

I have used the default config. The training set should be the combination of the training set and test it on dev clean or test clean, sorry that I have forgotten the details. I have tried different optimizers like Adam and SGD