DeepSpeech2

Python3 installation and complete setup only for prediction of Baidu's deep speech 2 model.

Implementation of DeepSpeech2 architecture for ASR. It is an open-source implementation of end-to-end Automatic Speech Recognition (ASR) engine, based on Baidu's Deep Speech 2 paper, with PaddlePaddle platform. Biadu’s pre-trained model for English is used for inference.

System Requirements:

OS: Ubuntu 16.04.5 LTS Language: Python3 and Bash Database: None

Tools:

PaddlePaaddle SoundFile Efficient Signal Resampling Python Speech Feature extraction

Packages:

List of Python packages has been attached in requirements.txt

Environment setup

Folders setup

sh folder_setup.sh

OR

git clone https://github.com/PaddlePaddle/DeepSpeech/ cd DeepSpeech/ mkdir checkpoints cd checkpoints/ mkdir baidu cd baidu/ mkdir step_final/ cd step_final/ cd .. cd .. cd .. cd models/ cd baidu_en8k/ sh download_model.sh cd .. cd .. cp -p models/baidu_en8k/params.pdparams checkpoints/baidu/step_final/params.pdparams cp -p ../data_utils/audio.py ../DeepSpeech/data_utils/audio.py cp -p ../data_utils/data.py ../DeepSpeech/data_utils/data.py cp -p ../data_utils/utility.py ../DeepSpeech/data_utils/utility.py cp -p ../model_utils/model.py ../DeepSpeech/model_utils/model.py cp -p ../utils/utility.py ../DeepSpeech/utils/utility.py cp -p ../utils/error_rate.py ../DeepSpeech/utils/error_rate.py cp -p ../decoders/swig_wrapper.py ../DeepSpeech/decoders/swig_wrapper.py cp -p ../swig_decoders.py ../DeepSpeech/swig_decoders.py cp -p ../_swig_decoders.py ../DeepSpeech/_swig_decoders.py cp -p ../_swig_decoders.cpython-35m-x86_64-linux-gnu.so ../DeepSpeech/_swig_decoders.cpython-35m-x86_64-linux-gnu.so cp -p ../create_manifest.py ../DeepSpeech/create_manifest.py cp -p ../infer_deepspeech2_baidu.py ../DeepSpeech/infer_deepspeech2_baidu.py cp -p ../models/lm/download_lm_en.sh ../DeepSpeech/models/lm/download_lm_en.sh cp -p ../deploy/demo_server.py ../DeepSpeech/deploy/demo_server.py cp -p ../deploy/demo_client.py ../DeepSpeech/deploy/demo_client.py mkdir test_dataset cp -r ../vaibhav_data_test_wav16k/ ../DeepSpeech/test_dataset/vaibhav_data_test_wav16k/ cd data mkdir baidu_en8k/ cd .. cd models/lm sh download_lm_en.sh cd ../..

Utilities setup

Either run required_packages.sh

OR

sudo apt-get install python3-pip sudo python3 -m pip install paddlepaddle

Libraries setup

sudo python3 -m pip install -r requirements.txt

OR

sudo python3 -m pip install scipy==1.2.1 sudo python3 -m pip install SoundFile==0.9.0.post1 sudo python3 -m pip install resampy==0.1.5 sudo python3 -m pip install python_speech_features

Other steps

Replace </data_utils/utility.py> from repo