Student Computer Science (f/m/d) - Integrating LLM into STT for output correction

Deutsches Zentrum für Luft-und Raumfahrt e.V.; Braunschweig; Vor Ort; EUR 40.000 - 80.000; Vollzeit

Zusammenfassung

An established industry player is seeking a motivated individual for a master's thesis project focused on enhancing Speech-to-Text (STT) models in aerospace engineering. This exciting opportunity involves conducting a literature review, implementing a demonstration, and evaluating STT performance in technical discussions. Candidates should possess strong programming skills in Python, a basic understanding of neural networks, and familiarity with deep learning frameworks. Join a collaborative environment where your contributions will drive innovation and improve technical communication in Concurrent Engineering Facilities. If you are passionate about technology and eager to tackle real-world challenges, this role is perfect for you.

Qualifikationen

• Master's thesis focusing on STT models in aerospace.
• Experience in Python and basic understanding of neural networks required.

Aufgaben

• Conduct literature review on STT models in technical environments.
• Develop a demonstration to test STT model performance.

Kenntnisse

Python

Neural Networks

Pandas

NumPy

Large Language Models

English Proficiency

Self-Motivation

Resilience

Ausbildung

Master's Degree

Tools

PyTorch

TensorFlow

Keras

Jobbeschreibung

What To Expect

Concurrent Engineering (CE) allows different disciplines—structural design, avionics, software—to work in parallel, making collaboration essential. Engineers must balance trade-offs, ensure smooth integration, and communicate effectively. Large Language Models (LLMs) can support this process by analyzing discussions, refining documentation, and assisting with decision-making. Our Modeling and Simulation group is developing a system that combines Speech-to-Text (STT) and LLM technologies to capture and process technical discussions in Concurrent Engineering Facilities (CEFs).

Your tasks

Your master’s thesis will focus on:

1. Literature Review – Investigate how well STT models perform in technical environments, particularly in aerospace. Analyze their handling of technical jargon, common errors, and correction methods. Explore how LLMs and domain-specific dictionaries can improve accuracy.
2. Implementation – Develop a small-scale demonstration to test an STT model in a simulated technical discussion. Evaluate its performance, identify limitations, and integrate LLM-based correction mechanisms to enhance transcription quality.

Your profile

• Proficiency in Programming Languages: Experience in Python is essential.
• Knowledge of Neural Networks: Basic understanding of neural networks and familiarity with deep learning frameworks such as PyTorch, TensorFlow, and Keras.
• Data Manipulation Skills: Experience with Pandas and NumPy for data analysis and manipulation.
• Familiarity with Large Language Models (LLMs): Basic understanding of LLM architectures, prompt engineering, and common challenges with their application.
• Strong English Proficiency: Good communication skills, both written and spoken.
• Self-Motivation and Independence: Ability to work independently, take ownership of tasks, and solve problems proactively.
• Resilience and Risk-Taking: Willingness to experiment, learn from failures, and iterate on solutions.