Towards a Scalable and Privacy-Preserving Audio Surveillance System
Bibliographic record
Abstract
The human voice is one of the passive biometrics that can be used in a surveillance system to uniquely identify individuals. It allows law enforcement agencies to detect and track suspects by deploying capturing devices (such as microphones) within a certain region. To address the clear privacy concerns of such an approach, we propose an efficient way of detecting suspects in public areas—through their voices—while preserving the privacy of innocent individuals. More precisely, our approach is quite suitable for large-scale surveillance systems, where millions of recordings are analyzed every day. Our privacy-preserving model is built on top of the most accurate speaker recognition systems, and we show that the accuracy loss due to the added privacy-preserving layer is negligible. The latter employs a highly efficient cryptosystem to securely compute the similarity scores between the captured utterances and the ones stored in the suspects' database. Specifically, the system computes, for each suspect, the encrypted Probabilistic Linear Discriminant Analysis (PLDA) score and obliviously matches it against a set threshold. More importantly, we show that our computation and communication overhead is significantly lower compared to the state-of-the-art techniques, which facilitates a real-time surveillance operation. Our protocol necessitates a single round of communication between the server and the capturing device and, for a database of 100 suspects, the online computation time is only 135 ms on the capturing device and 35 ms on the server, whereas the required communication is 12 KB.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.003 |
| Open science | 0.002 | 0.004 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".