A TensorFlow implementation of Baidu's DeepSpeech architecture

DeepSpeech is an open source Speech-To-Text engine, using a model trained by machine learning techniques based on Baidu's Deep Speech research paper <https://arxiv.org/abs/1412.5567>. Project DeepSpeech uses Google's TensorFlow <https://www.tensorflow.org/> to make the implementation easier.

Training Your Own Model

https://github.com/mozilla/DeepSpeech/blob/master/doc/TRAINING.rst#training-your-own-model

'Turkish Language TSV file' https://voice.mozilla.org/tr/datasets

Installation

git clone https://github.com/mozilla/DeepSpeech
cd DeepSpeech
pip3 install -r requirements.txt
pip3 install deepspeech

To install and use deepspeech all you have to do is:

pip3 install deepspeech

pre-trained Turkish model

For the language model, I used kenlm’ lmplz -o 2 < vocabulary > text.arpa build_binary text.arpa lm.binary

after training

loss = 6.42

/model/output_graph.pb

Training model

sudo ./run-ldc93s1.sh

Quicker inference can be performed using a supported NVIDIA GPU on Linux. See the `release notes <https://github.com/mozilla/DeepSpeech/releases/latest>`_ to find which GPUs are supported. To run ``deepspeech`` on a GPU, install the GPU specific package:

Install DeepSpeech CUDA enabled package

pip3 install deepspeech-gpu

Testing model

download lm.binary file from google drive

https://drive.google.com/open?id=1n2VCKosd2JsCVF1TQWIkKbVdeLQf2OYJ

deepspeech --model '/model/output_graph.pb' --lm '/data/lm/lm.binary' --trie '/data/lm/trie' --audio example.wav

Real-time DeepSpeech Analysis

python code example

https://discourse.mozilla.org/t/real-time-deepspeech-analysis-using-built-in-microphone/42669