Computer Audio

Deemsoft is proud to state that we can produce premium AI technologies such as Machine Learning/Deep Learning for Audio processing. Below case study developed for our client Audiosinc provides more details about human voice recognition, speaker recognition, language translation, natural language processing, etc.


Problem: There is tremendous advancement in AI specific to Automatic Speech Recognition(ASR) and Natural Language Processing still significant inaccuracies in STEM related subjects and specifically Mathematics and Physics is not even acceptable level. 

Objective: Increase accuracy in STEM subject lectures like Biology.

Solution: We collected around 5000 hours open source stem related content from various universities and youtube lectures.  Most of this kind of data is not labeled and we used some automation and manual effort labeled the data. Using TensorFlow deep learning platform and LSTM(Long Short Term Memory) model to train the data. We trained on top of DeepSpeech model. Results were promising and continued for another 5000 hours with data augmentation techniques and finaly we were able to produce better results compared to IBM Watson Speech2Text model and Google cloud Speech-to-text at the time of our test. 

Result WER: 19% our model where as IBM 26% on 1000 hours of test data



Cloud Instance: AWS EC2

GPU: V100 Nvidia GPU 

Platform: DeepSpeech V 0.90

Technology: TensorFlow GPU 1.514 

Model : LSTM

Corpus: 5000 hours original and 5000 hours augmented

Computer Vision

Computer Vision: Deemsoft is exited to announce that we have developed cutting edge technology using AI technologies like Machine Learning and Deep Learning. Developed AI model to read images and interpret such face recognition, object recognition, X-Ray analyzer, etc. Below example developed for AIMS demonstrates practical application.


  • Missed and incorrect diagnosis of chest X-Ray.
  • Higher wait times due to Radiologist constraints

Objective: Increase accuracy in reading Human chest X-Ray.


From: Our X-Ray Analytics solution   Utilized ChexNet data set & Stanford University DenseNet  Neural network model 

To  Analyze X-Rays & provide   potential issue description with % probability


  • It helps hospitals serve patients faster with higher accuracy
  • 30% increase in savings by reducing errors and delays


Cloud Instance: AWS EC2

GPU: V100 Nvidia GPU 

Platform: Stanford Densenet Model

Technology: TensorFlow GPU 1.514 

AI Model : DenseNet/CNN

Corpus: Chest x-ray dataset from National Institue of Health(NIH) US Gov.


Deemsoft demonstrated the real life AI applications in data classification, data analytics, data predictions and data Intelligence. Below section describes what we have implemented for Federal Co Operative Bank client.


  • High volumes of loan processing
  • Complex financial documents and land records

Objective: Increase accuracy in loan approval process.


Using digital loan application details and OCR produced collateral documents as input to our system. Developed  linear regression and logistic regression model using python sklearn, nltk, imblearn, pondas,  Once application ready then can submit to our  fraud analytics system which intern produces the risk analysis report and identifies the reasons for concern.


  • It helps banks process applications faster with higher accuracy
  • 10% increase in savings while avoiding defaults.


Cloud Instance: AWS EC2

GPU: V100 Nvidia GPU 

Platform: Python

Technology: IMBLearn, SKLearn, NLTK

AI Model : Linear Regression & Logical Regression

Corpus: gensim

For additional details or any other use case feel free to contact us

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>