A software using MLP (MultiLayer Perceptron) neural network,NLP and speech to text that can learn and evaluate past salesperson's speech patterns and their effect on customers' purchasing decisions and predict customers' purchasing decision from sales rep's speech.
- Support speech to text allow companies to analyze sales rep's speech by simply recording it;
- Customized AI model users can train their AI model to make the prediction more accurate by fitting the industry;
- Pretrained model A pretrained model is also given for general use;
- Usage across different devices The model is stored on Google drive and can be loaded from different devices by using Google accounts having access to the same folder;
The Windows executable file can be downloaded here:
https://drive.google.com/drive/folders/1bmHq8j93-mv87ilJEhM17VJnQAVje5HQ?usp=sharing.
It is built using PyInstaller with main.spec.
Clone this repo with git clone https://github.com/DanielSinTY/salesTrainingProgram.git or using GitHub desktop.
All required modules and packages are listed in requirements.txt. Use pip install -r requirements.txt to install all the packages and modules needed. You may want to use a virtual environment for this.
Follow the instruction here (https://pythonhosted.org/PyDrive/quickstart.html#authentication) to create an API to be used by PyDrive to connect to Google Drive. Remember to rename the downloaded json file as client_secrets.json and place it in the working directory.
2. Authetication for Google Drive
A Google Drive Folder named model has to be created to store the model before using any of the functions. model.joblib in this repo can also be put in the folder for STEP to use the pretrained model.
Open the .exe file or run main.py in /salesTraining and STEP will open your browser and prompt you to sign in to Google. Sign in using an account with access to the folder model and authorize the software. You can close the browser after the authetication flow has been completed (this will be shown in the browser).
After authetication, the GUI of STEP will pop up. Click Analyse to predict customers' purchasing decision from a sales rep's speech. Click Train to train the AI model with previous sales reps' speech. Click Delete to delete the AI model.
By clicking Analyse, you can either record a sales speech or upload an audio file of recording to predict customers' purchasing decision based on the speech.
After recording a speech of uploading an audio file, you can listen to the recording, restart recording or uploading, or start analysis on the sales speech.
By clicking Start analysing, STEP will load the model from Google Drive, extract features from the speech, and predict customers' purchasing decisions with the model based on the speech. After analysing, either of two outputs will be shown. 
By clicking Train, STEP will prompt you to upload files of previous successful (i.e. the customer purchased the product after that) and poor (i.e. the customer did not purchase) sales speech respectively to be used as data to train the MLP model. You can either upload audio recordings in .wav or transcript in .txt.
After uploading all the files, click Start training and STEP will start training the MLP model with the files uploaded. STEP will first extract features from all the files, and then feed into the MLP model to train it. Progress bar of feature extraction and iteration number during feeding will be shown on the screen. The process may take a few minutes to couple of hours depending on size of dataset (number of files uploaded) and computing power. If a model exists in the Google Drive, STEP will train that model using the new files uploaded. A new MLP model will be created otherwise.
When you see 
model and you can use it to analyse sales speech.
By clicking Delete, a warning box will be shown before deleting the model. Click Yes to delete the model from the Google Drive folder.
After successfully deleting the model, another info box will be shown. You can train a new model after that.
The following modules are used in STEP:
- scikit-learn to train and use the MLP model
- joblib to store and load MLP model
- pydub to manipulate audio
- SpeechRecognition to connect to Google speech recognition API and perform speech to text transcription
- Empath to analyse and categorize the words in speech
- SpaCy to analyse sentence patterns
- PyDrive to connect to Google Drive
- soundDevice and soundFile to record audio
- numpy to manipulate array
- pygame to play the recording
- Tkinter to build the UI
- PyInstaller to package the program into exe