This work was based on the code developed by tensorflow
This is an implementation of the DeepSpeech2 model. Current implementation is based on the code from the authors' DeepSpeech code and the implementation in the MLPerf Repo. DeepSpeech2 is an end-to-end deep neural network for automatic speech recognition (ASR). It consists of 2 convolutional layers, 5 bidirectional RNN layers and a fully connected layer. The feature in use is linear spectrogram extracted from audio input. The network uses Connectionist Temporal Classification CTC as the loss function.
The OpenSLR LibriSpeech Corpus are used for model training and evaluation.
The training data is a combination of train-clean-100 and train-clean-360 (~130k examples in total). The validation set is dev-clean which has 2.7K lines. The download script will preprocess the data into three columns: wav_filename, wav_filesize, transcript. data/dataset.py will parse the csv file and build a tf.data.Dataset object to feed data. Within each epoch (except for the first if sortagrad is enabled), the training data will be shuffled batch-wise.
Add the top-level /models folder to the Python path with the command:
export PYTHONPATH="$PYTHONPATH:/path/to/models"
If you encounter this error -- AttributeError: module 'tensorflow.python.estimator.estimator_lib' has no attribute 'SessionRunHook', you can get TensorFlow official models release 1.1 to replace the models.
First install shared dependencies before running the code. Issue the following command:
pip3 install -r requirements.txt
or
pip install -r requirements.txt
and
sudo apt-get install sox
To download the dataset, issue the following command:
python data/download.py
Arguments:
--data_dir: Directory where to download and save the preprocessed data. By default, it is/tmp/librispeech_data.
Use the --help or -h flag to get a full list of possible arguments.
To train and evaluate the model, issue the following command:
python deep_speech.py
Arguments:
--model_dir: Directory to save model training checkpoints. By default, it is/tmp/deep_speech_model/.--train_data_dir: Directory of the training dataset.--eval_data_dir: Directory of the evaluation dataset.
There are other arguments about DeepSpeech2 model and training/evaluation process. Use the --help or -h flag to get a full list of possible a
rguments with detailed descriptions.
A shell script run_deep_speech.sh is provided to run the whole pipeline with default parameters. Issue the following comman d to run the benchmark:
Note by default, the training dataset in the benchmark include train-clean-100, train-clean-360 and train-other-500, and the evaluation dataset i nclude dev-clean and dev-other.