CODEWITHSUNDEEP
androidtensorflowspeech-recognition
Oke Uwechue
Oke Uwechue

Reputation: 421

What's the most effective way to implement an offline continuous speech recognizer on Android using Tensorflow?

I'm looking for a way to implement an offline continuous speech recognizer running on Android and preferably using the Tensorflow engine.

The app needs to be able to recognize any word out of a predefined set of words in a continuous stream of speech and the assumption is that the Android device may be in an area that does not support an internet connection, so Google services and such are out of the question.

Are there perhaps any RNN or CNN models that can achieve this?

Upvotes: 1

Views: 877

Answers (1)

Shubham Panchal
Shubham Panchal

Reputation: 4289

You can use the Speech Recognition ( scroll down and search for "Speech Recognition" ) model available on TensorFlow Lite. It can generate probability scores for words spoken in an audio stream. The best thing is that they have provided an Android sample (the example is now deprecated, but the last available commit is here).

Using a TensorFlow Lite model, you will not require an active internet connection to make predictions. This can compromise the app's size ( the .tflite model file could have a large size though ).

You can build a simple method which extracts necessary words from the model's predictions.

Upvotes: 1

Related Questions