A Telecommunication Provider company receives thousands of calls per day and needs to improve its customer service by verifying the customers in recorded calls. Thus, the calls are recorded and saved for further evaluation by the Quality Control department of the company. A part of the evaluation process is the verification of the customer which is a difficult task in the case this is not done by the call operator during the call. The application being looked for will provide the staff of the Quality Control department with the opportunity to load recorded calls and automatically verify the customers. The application will help the company to decrease the time needed for the evaluation process and improve its services, increase customers and revenue.

We decided to use semi-structured interviews, which involves both fixed and opened questions. Compared to questionnaires, interviewing allows us to guide the interviewees to provide more suitable information we need, especially the answers to opened questions. 

Our previous project topic was "Speaker verification: Find out if your voice sounds the same as your favourite celebrities". After we finished the HCI coursework, the project topic was changed to "Speaker verification in phone calls". We didn't have enough time to do personas for the current project. So, the personas below are off-topic.

The goal of this project is to develop an innovative speech recognition methods using machine learning algorithms. Before implementing the state-of-the-art algorithm to verify the voice, literature review on the current practices would be conducted. According to the comparisons between various type of algorithms, a deep learning model would be set up by utilising open source libraries and trained based on any voice dataset. The learning algorithm would be demonstrated as a speaker verification application via webpages.

We understood our aim for the project. This "Won't have" part will be for things that we will agree or disagree on doing in the near future in order to be clear on what our prototype can do.

ID	Requirement	Priority
1	Testing for finding out the framework with the highest accuracy	Must have
2	A "Upload Data" page	Must have
3	A "Start Verifying" page	Must have
4	A "Home" page	Must have
5	A "Processing" page	Must have
6	A "User Verified" page	Must have
7	A "Submit" page	Must have
8	A "Consent" Page	Must have
9	Literature review	Must have
10	Comparisons between different feature extractions	Must have
11	Comparisons between dimensionality reduction	Must have
12	Comparisons between diufferent learning models	Must have
13	Database	Must have
14	File Uploader	Must have
15	Audio Recorder	Must have
16	Feature extraction: MFCC	Must have
17	Modeling: HMM	Must have
18	Modeling: GMM	Must have
19	Front-End: Modify with AJAX	Must have
20	Use Existing Code (libraries and APIs)	Must have
21	Comparisons between different frameworks	Should have
22	Speaker Recognition VS Biometric Feature Extractions	Should have
23	A "Recent verifications" page	Could have
24	Feature extraction: PLP	Could have
25	Feature extraction: LPC	Could have
26	Optimisation - Check 2 or 3 digits of the number to find the user faster	Could have
27	{To be added}	Won't have

ID	User Case For User
UCU1	Uploading data and training models
UCU2	Verifying

---

User Case
ID	UCU1
Actor	User
Description	Uploading data and training models
Main Flow	1. Click "Upload Data" button 2. Upload audio files by file chooser or drag & drop and click "Upload" button 3. Select who is the speaker in the audio files 4. Check the checkbox about user terms and click "Start" button.
Result	Model of certain speaker stored

---

User Case
ID	UCU2
Actor	User
Description	Verifying
Main Flow	1. Click "Start Verifying" button 2. Upload an audio file by file chooser or drag & drop and click "Upload" button 3. Select the speaker to be verified 4. Check the checkbox about user terms and click "Verify audio" button.
Result	Verification result shown