Master Thesis – Core Tech Media

Audio processing on embedded devices requires a balance between performance, energy use, privacy and limited computing resources. Axis Communications offers four alternative master's thesis topics covering audio classification and speaker anonymization for cameras, battery-powered devices and intercoms. The projects aim to investigate methods that make advanced audio capabilities practical on resource-constrained hardware.

The assignment

• Choose one of four research directions, rather than completing all four.
• Knowledge distillation: investigate whether large computer-based models can transfer knowledge to compact audio classifiers and improve their performance on camera hardware.
• Synthetic training data: compare physical simulation, soundscape mixing and generative audio models for rare sound-event classification. Evaluate generalization to real recordings, the amount of real training data still required and whether Fréchet Audio Distance predicts classification performance. Develop a reproducible generation and evaluation pipeline.
• Energy-efficient classification: assess cascaded detection, model compression, cheaper audio features and context-dependent operation. Benchmark detection accuracy against measured energy consumption on embedded hardware, identify energy costs across the pipeline and prototype an adaptive system.
• Speaker anonymization: compare lightweight approaches with VoicePrivacy baselines in terms of identity protection and intelligibility. Test latency, memory and computing requirements on embedded hardware, including robustness against attackers who know the anonymization method. Optionally investigate the effects of codecs, noise, reverberation and restricted audio bandwidth.

Who we are looking for

• Master's students in computer science, applied mathematics, engineering physics, electronics engineering or a related discipline.
• Affiliation with a Swedish university or university college, either as a regular student or through an exchange programme.
• Relevant coursework in C/Python and machine learning or AI for the knowledge-distillation, synthetic-data and speaker-anonymization topics.
• Signal-processing coursework is also relevant for the speaker-anonymization topic.
• Two students applying together.

What we offer

• A choice of four research proposals connected to audio technology in Axis products.
• An international technology company with an open, friendly and collaborative working culture.

Practical details

Location: Lund. Scope: two master's students, with 30 credits per student. Apply through the thesis advertisement by 27 November 2026; applications may be written in Swedish or English. Submit separate applications and identify your thesis partner in each application. Include your CV and university transcript. Questions: Daniel Andersson, Core Technologies Media Manager, Daniel.B.Andersson@axis.com.