CODEWITHSUNDEEP
pythontensorflowspeech-recognitionconv-neural-network
Shivang Sharma
Shivang Sharma

Reputation: 76

Tensorflow joining two models

Suppose I have created a speech to text model 'A' which recognizes 20 speech commands and model 'B' which recognizes other 20 commands .Is there a way so that i can join both models so that it recognizes total 40 words.

Upvotes: 2

Views: 239

Answers (1)

csteel
csteel

Reputation: 468

I think it depends on your model. If you are using a deep learning model like and LSTM or RNN, and depending on what libraries you use, it might be possible. Many people use pre-trained CNNs to then train their image recognition and generation models. I am not sure how you would join the weights of the two already trained models. Starting from scratch, and using Keras I would train one model, save the weights, load the weights into the other model (assuming they have the same architecture and parameters, and then train the second model on the other 20 words. Also, it might be worth looking into some of the distributed learning solutions out there. I only know of vendor with products (no open source that I have heard of) that offer products that centralize model weights and allow you train data on different data in different locations and then merge those weights centrally. For instance, different hospitals can share model training without sharing data, which is very beneficial when you think about privacy.

Upvotes: 1

Related Questions