GitPlanet

Cheap and reliable Node.js hosting starts at $3/month, and $1/month static HTML hosting

Created with love in Canada, visit hostnodejs.com today

Feel like to post an Ad? Learn Details
All Projects → mravanelli → Theano Kaldi Rnn

mravanelli / Theano Kaldi Rnn

THEANO-KALDI-RNNs is a project implementing various Recurrent Neural Networks (RNNs) for RNN-HMM speech recognition. The Theano Code is coupled with the Kaldi decoder.

Programming Languages

perl
6916 projects

Labels

deep-learningdeep-neural-networksrnnrecurrent-neural-networkstheanokaldigru

Projects that are alternatives of or similar to Theano Kaldi Rnn

Pytorch Kaldi
pytorch-kaldi is a project for developing state-of-the-art DNN/RNN hybrid speech recognition systems. The DNN part is managed by pytorch, while feature extraction, label computation, and decoding are performed with the kaldi toolkit.
Stars: ✭ 2,097 (+6664.52%)
Mutual labels:  deep-neural-networks, rnn, recurrent-neural-networks, kaldi, gru
Rnn ctc
Recurrent Neural Network and Long Short Term Memory (LSTM) with Connectionist Temporal Classification implemented in Theano. Includes a Toy training example.
Stars: ✭ 220 (+609.68%)
Mutual labels:  rnn, recurrent-neural-networks, gru, theano
Easy Deep Learning With Keras
Keras tutorial for beginners (using TF backend)
Stars: ✭ 367 (+1083.87%)
Mutual labels:  deep-neural-networks, rnn, gru
Rmdl
RMDL: Random Multimodel Deep Learning for Classification
Stars: ✭ 375 (+1109.68%)
Mutual labels:  deep-neural-networks, rnn, recurrent-neural-networks
Gru Svm
[ICMLC 2018] A Neural Network Architecture Combining Gated Recurrent Unit (GRU) and Support Vector Machine (SVM) for Intrusion Detection
Stars: ✭ 76 (+145.16%)
Mutual labels:  rnn, recurrent-neural-networks, gru
Deepseqslam
The Official Deep Learning Framework for Route-based Place Recognition
Stars: ✭ 49 (+58.06%)
Mutual labels:  deep-neural-networks, rnn, recurrent-neural-networks
Gdax Orderbook Ml
Application of machine learning to the Coinbase (GDAX) orderbook
Stars: ✭ 60 (+93.55%)
Mutual labels:  deep-neural-networks, recurrent-neural-networks, gru
Pytorch Learners Tutorial
PyTorch tutorial for learners
Stars: ✭ 97 (+212.9%)
Mutual labels:  deep-neural-networks, rnn, recurrent-neural-networks
Bitcoin Price Prediction Using Lstm
Bitcoin price Prediction ( Time Series ) using LSTM Recurrent neural network
Stars: ✭ 67 (+116.13%)
Mutual labels:  deep-neural-networks, rnn, recurrent-neural-networks
sequence-rnn-py
Sequence analyzing using Recurrent Neural Networks (RNN) based on Keras
Stars: ✭ 28 (-9.68%)
Mutual labels:  theano, recurrent-neural-networks, rnn
theano-recurrence
Recurrent Neural Networks (RNN, GRU, LSTM) and their Bidirectional versions (BiRNN, BiGRU, BiLSTM) for word & character level language modelling in Theano
Stars: ✭ 40 (+29.03%)
Mutual labels:  theano, gru, rnn
Agentnet
Deep Reinforcement Learning library for humans
Stars: ✭ 298 (+861.29%)
Mutual labels:  deep-neural-networks, theano
Theano lstm
🔬 Nano size Theano LSTM module
Stars: ✭ 310 (+900%)
Mutual labels:  gru, theano
Returnn
The RWTH extensible training framework for universal recurrent neural networks
Stars: ✭ 290 (+835.48%)
Mutual labels:  recurrent-neural-networks, theano
First Steps Towards Deep Learning
This is an open sourced book on deep learning.
Stars: ✭ 376 (+1112.9%)
Mutual labels:  deep-neural-networks, recurrent-neural-networks
Tensorflow Char Rnn
Char-RNN implemented using TensorFlow.
Stars: ✭ 429 (+1283.87%)
Mutual labels:  rnn, recurrent-neural-networks
Nmt Keras
Neural Machine Translation with Keras
Stars: ✭ 501 (+1516.13%)
Mutual labels:  gru, theano
Ner Lstm
Named Entity Recognition using multilayered bidirectional LSTM
Stars: ✭ 532 (+1616.13%)
Mutual labels:  deep-neural-networks, recurrent-neural-networks
Rnnsharp
RNNSharp is a toolkit of deep recurrent neural network which is widely used for many different kinds of tasks, such as sequence labeling, sequence-to-sequence and so on. It's written by C# language and based on .NET framework 4.6 or above versions. RNNSharp supports many different types of networks, such as forward and bi-directional network, sequence-to-sequence network, and different types of layers, such as LSTM, Softmax, sampled Softmax and others.
Stars: ✭ 277 (+793.55%)
Mutual labels:  rnn, recurrent-neural-networks
Punctuator2
A bidirectional recurrent neural network model with attention mechanism for restoring missing punctuation in unsegmented text
Stars: ✭ 483 (+1458.06%)
Mutual labels:  recurrent-neural-networks, theano
View All Similar Projects ➔

Introduction:

THEANO-KALDI-RNNs is a software which offers the possibility to use various Recurrent Neural Networks (RNNs) in the context of a Kaldi-based hybrid HMM/RNN speech recognizer.

Note: A new project called "pytorch-kaldi" https://github.com/mravanelli/pytorch-kaldi is now available. If you are interested, please take a look into it.

The current version supports the following standard architectures:

  • reluRNN
  • LSTM
  • GRU

The code also considers some architectural variations:

  • ReluGRU
  • Minimal Gated Recurrent Units (M_GRU)
  • M-reluGRU (also known as Light-GRU)

The latter architectures have been explored in [1] (see reference). Please cite this paper if you use this toolkit or a part of it.

All the RNNs are based on a state-of-the-art technology which includes:

  • Bidirectional architectures
  • Bach Normalization (applied to feed-forward connections)
  • Recurrent Dropout
  • RMSE prop optimization

Prerequisites:

  • If not already done, install KALDI (http://kaldi-asr.org/) and make sure that your KALDI installation is working.

  • Run the original TIMIT kaldi recipe in egs/timit/s5/run.sh and check whether everything is properly working. This step is necessary to compute features and labels that will be inherited in the theano/python part of this code.

  • Install THEANO (http://deeplearning.net/software/theano/install.html) and make sure your installation is working. Try for instance to type import theano in the python environment and check whether everything works fine.

The code has been tested with:

  • Python 2.7
  • Ubuntu 14 and RedHat (6,7)
  • Theano 0.8.1
  • Bash shell

How to run an experiment:

1. Run the Kaldi s5 baseline of TIMIT.

This step is necessary to derive the labels later used to train the RNN. In particular:

  • go to $KALDI_ROOT/egs/timit/s5.
  • run the script run.sh. Make sure everything (especially the tri3-ali part) works fine. Note that the s5 recipe computes tri3-ali for training data only. Please, computed them for test and dev data as well with the following commands:
steps/align_fmllr.sh --nj "$train_nj" --cmd "$train_cmd" \
 data/dev data/lang exp/tri3 exp/tri3_ali_dev

steps/align_fmllr.sh --nj 24 --cmd "$train_cmd" \
 data/test data/lang exp/tri3 exp/tri3_ali_test

2. Convert kaldi features and labels into pkl format.

  • Set your own paths in compute_features.sh (ali_dir,ali_dir_dev,ali_dir_test,data_dir,...)
  • Run compute_features.sh.

3. Write the Config file.

  • Open the file TIMIT_GRU.cfg and modify it according to your paths. Feel free to modify the DNN architecture and the other optimization parameters according to your needs. See the comments in the TIMIT_GRU.cfg file for a brief description of each parameter. The number of outputs N_out can be found with the following kaldi command (see number of pdfs):
am-info exp/tri3/final.mdl
  • The required count_file in the config file (used to normalize the DNN posteriors before feeding the decoder) corresponds to the following file:

/exp/dnn4_pretrain-dbn_dnn/ali_train_pdf.counts

4. Run the experiment.

  • Open the file run_exp.sh. Please note that the parameters defined in run_exp.sh have the priority over the parameters specified in the cfg_file. For example, the learning rate defined in run_exp.sh is the one that will be actually used during training.
  • Set cfg_file, graph_dir, data_dir, ali_dir according to your specific paths
  • To replicate GRU experiments of the paper [1], set cfg_file=TIMIT_GRU.cfg in run_exp.sh
  • To replicate M_reluGRU experiments (improved architecture) of the paper [1], set cfg_file=TIMIT_M_reluGRU.cfg in run_exp.sh

5. Check the results.

  • After training, forward and decoding phases are finished, you can go into the kaldi_decoding_scripts foder and run ./RESULTS to check the system performance.

  • Note that the performance obtained can be slightly different from that reported in the paper due, for instance, to the randomness introduced by different initializations. To mitigate this source of randomness and perform a fair comparison across the various architectures, in [1] we ran more experiments with different seeds (i.e., setting a different seed in the cfg_file) and we averaged the obtained error rates.

Please, note that this is an ongoing project. It would be helpful to report us any issue!

Reference:

[1] M. Ravanelli, P. Brakel, M. Omologo, Y. Bengio, "Improving speech recognition by revising Gated Recurrent Units", in Proceedings of Interspeech 2017

https://arxiv.org/abs/1710.00641

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].