SpeechRecognition

Speech recognition system implemented using TensorFlow. More explanation in Medium article: https://aruno14.medium.com/a-journey-to-speech-recognition-using-tensorflow-1fc1169fef99

Used libraries

pip install tensorflow tensorflow_io

Files explanation

test_load.py

Load labels file of Mozilla Common Voice: https://commonvoice.mozilla.org/en/datasets

test_lstm.py

Simple LSTM model to predict next words of a sequence of words. It uses Mozilla Common Voice dataset labels file.

test_trad.py

Seq2Seq model to translate sentence in same language as the input. It uses Mozilla Common Voice dataset labels file.

test_words.py

Simple LSTM model to convert audio into word. It uses Speech command Dataset: https://ai.googleblog.com/2017/08/launching-speech-commands-dataset.html

test_wordsFr.py

Simple LSTM model to convert audio into French word. It uses a self-made Dataset included in this repository.

test_words_compare.py

Files to compare accuracy of different audio representation. More explanation in Medium article: https://aruno14.medium.com/comparaison-of-audio-representation-in-tensorflow-b6c33a83d77f

test_words_human.py

Same as test_words.py, but high frequencies are cut.

test_voice.py

Read audio input data from mic and predict word in real time.

server_express.js

Simple WebServer in NodeJS.

convert_mp3towav.sh

Convert all mp3 files of current folder in wav.

sentence.py

Seq2Seq model to convert audio sentence in text. It uses Mozilla Common Voice English dataset.

sentence_fr.py

Seq2Seq model to convert audio sentence in text. It uses Mozilla Common Voice French dataset.

sentence_jp.py

Seq2Seq model to convert audio sentence in text. It uses Mozilla Common Voice Japanese dataset. It uses MeCab for word segmentation.

HTML pages

In order to avoid CORS probleme, we use a local WebServer. You can use python3 -m http.server 3000 or node server_express.js and access: http://localhost:3000.

Model can recognize: ['down', 'go', 'left', 'no', 'right', 'stop', 'up', 'yes']

In addition to English, there is also a French test page: http://localhost:3000/indexFr.html.

Model can recognize: ['bonjour', 'salut', 'merci', 'aurevoir']

Below command has been used to convert model to TensorFlow.js:

~/.local/bin/tensorflowjs_converter model_reco/ quantized_model/ --input_format tf_saved_model --output_format tfjs_graph_model --quantize_float16