**NOTEBOOK-LM Link for guys that is Too long-don’t Read: https://notebooklm.google.com/notebook/ef0af5ae-bc34-454c-b78b-65fa7ab53107**

→ In this lesson, we shall cover the concept of Loss & Optimization Goal in training a MDD-model (which essential in the process of model fine-tuning process)

→ This session shall include these 3 types of loss:

1. CTC-Loss
2. Cross-Entropy-Family Loss (on the Frame-Level knowledge)
3. Weighted-Loss (a combination of 1 & 2)
4. When should use CTC & When should use Cross-Entropy

Definition: Loss function is a mathematic function that use to track the amount of error that your machine-learning model been made - So lower the loss, the better your model - (Bergmann & Stryker, 2024)

In the development of speech-processing model, 3 types of loss function that you shall encounter much, and we shall explain the idea behind them, why it’s matter and when you should use them.

Audio-podcast version

1 - CTC loss most basic loss function - based on (Graves et al., 2006)

2 - Cross-Entropy-Family Loss (on the Frame-Level knowledge)

3 - Weighted-Loss - combination of 1 and 2

4. When should use which - refer to this (Biswajit, 2024)

Reference for Lesson 3:

In the next session we shall cover metric that people may use for evaluation and quality assurance of the MDD system Lesson 4: Metrics & Evaluation in MDD-Problem