Voice-Driven Semantic Perception for UAV-Assisted Emergency Networks

Andr\'e Coelho; Nuno Saavedra; Pedro Ribeiro; Rui Campos

REVIEW 3 major objections 5 minor 20 references

Voice-Driven Semantic Perception for UAV-Assisted Emergency Networks

T0 review · 3 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read SIREN converts emergency radio traffic into structured, machine-readable information — locations, units, severity, and QoS needs — and its evaluation on synthetic scenarios argues this voice-driven perception is feasible for UAV-assisted ne

desk verdict SIREN is a sensible voice-to-semantics pipeline for UAV emergency networks, but the abstract oversells the evidence: all results come from five synthetic LLM/TTS runs per scenario, with no real radio data. read the letter →

arxiv 2602.17394 v2 pith:2P273I73 submitted 2026-02-19 cs.NI cs.AIcs.SD

Nuno Saavedra , Pedro Ribeiro , André Coelho , Rui Campos This is my paper

classification cs.NIcs.AIcs.SD

keywords SIRENUAV-assistedemergencynetworksautomaticspeechrecognitionlargelanguagemodelssemanticextractionspeakerdiarizationQoSexpectationvoicecommunications

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proposes SIREN, a modular pipeline that listens to emergency voice radio, transcribes it, and uses a large language model to pull out structured fields: which units are speaking, where they are, how severe the emergency is, and what bandwidth or video support they need. The authors' central claim is that this voice-driven extraction is feasible for UAV-assisted emergency networks, where drones act as relays and managers. They evaluate SIREN on synthetic emergency dialogues spanning two languages, clean and noisy audio, and two to six speakers, reporting that locations and units are reliably identified in most scenarios, while QoS-expectation extraction varies with how explicitly needs are stated. The main limits, they find, are speaker diarization when voices sound alike and geographic disambiguation during geocoding. If SIREN works as described, voice radio becomes a semantic sensor that could inform drone positioning and resource allocation.

What carries the argument

The load-bearing mechanism is the SIREN pipeline itself, a modular chain from audio to structured semantics: automatic speech recognition (with a cloud backend that also provides diarization and sentiment analysis, plus a lighter local model for offline use), information extraction by an LLM constrained by a strict JSON schema that forces explicit outputs for locations, units, emergency level, and QoS fields, and a deterministic validation layer using named-entity recognition to cross-check locations, speaker diarization to check unit attribution, and sentiment analysis to adjust severity. The pipeline emits a JSON object designed to plug into network management tools. The evaluation is carr

What would settle it

Run SIREN on a set of real emergency or incident radio recordings with known ground-truth transcripts, and measure location, unit, and QoS extraction accuracy. If word error rates climb far above the synthetic values or semantic extraction success drops markedly under realistic channel artifacts and speaker overlap, the paper's feasibility claim would need to be scaled back. A sharper test: use recordings with two acoustically similar real speakers; if diarization often assigns the wrong unit to a location or QoS request, the claimed reliability fails for multi-responder operations.

Watch

Extended reading notes

Core claim

On its own terms, the paper's discovery is that a three-stage pipeline — automatic speech recognition, schema-constrained LLM extraction, and deterministic NLP validation — can turn emergency radio conversations into a JSON structure containing locations, emergency level, responding units with location associations, and QoS expectations such as video uplink needs. In the paper's evaluation over five synthetic scenarios, location and unit identification succeed in nearly all runs, including on noisy audio, while speaker-count estimation fails when text-to-speech voices are acoustically similar and geocoding sometimes resolves place names to wrong coordinates. The paper concludes that voice-dr

Load-bearing premise

The results rest on the assumption that synthetic dialogues generated by an LLM, spoken by a text-to-speech engine, and degraded with added noise faithfully represent real first-responder radio traffic, including speaker overlap, channel compression, and spontaneous speech; if that assumption fails, the measured success rates may not transfer to real operations.

Editorial extensions

If this is right

Voice radio can serve as a perception input for UAV placement and resource-allocation decisions in emergency networks, without requiring visual line-of-sight.
Human operators can get a structured, map-rendered picture of who is where and what link support they need, enabling human-in-the-loop decision support.
Offline operation is possible with a local ASR model, but its noise robustness is materially worse than a cloud ASR API, so offline deployments would need more capable hardware or noise-robust front-ends.
The main bottlenecks are not transcription or semantic extraction but speaker attribution and geocoding; future work should constrain search spaces and use domain-specific location gazetteers.
The pipeline is domain-agnostic and could be applied to other voice-coordination settings such as search-and-rescue or surveillance.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Real first-responder radio traffic includes push-to-talk clipping, channel compression, overlapping transmissions, and emotional stress; synthetic text-to-speech testing likely underestimates how hard diarization and transcription will be, so the reported success rates should be read as upper bounds until tested on authentic recordings.
A failure in speaker diarization could silently misattribute a unit's location or QoS needs, which is acceptable for situational awareness but would be risky if the output feeds autonomous network management without a human check.
The schema-constrained prompting is the main customization point, so the same pipeline could be reprompted to extract other structured fields, such as casualty counts, medical needs, or equipment status, without redesigning the architecture.
The extracted QoS throughput values are expectations stated by callers, not measured link requirements; network managers should treat them as soft requests when allocating bandwidth.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Here’s the short version: SIREN is a modular pipeline that turns emergency voice traffic into structured JSON — locations, units, QoS needs — by chaining ASR, LLM extraction, and deterministic NLP checks for validation. As a proof-of-concept it’s coherent, and the integration is genuinely new for UAV-assisted emergency networks. The authors position it as a perception layer, not a network manager, which is the right scope.

The best parts: schema-constrained prompting with deterministic validation is a sane way to limit LLM hallucination, and the local/cloud ASR comparison is useful. The paper is honest about its limits — the discussion and future work explicitly name diarization, geocoding ambiguity, and the lack of authentic recordings. They also correctly distinguish extraction errors from geocoding errors.

The soft spots are real. Every success rate in Tables III and IV comes from five runs per scenario with no confidence intervals, and all audio is LLM-written dialogue spoken by ElevenLabs TTS with Python-added noise. That means no radio compression, no genuine overlap, no real channel artifacts. The 0% speaker-count results in Scenarios 2, 3, and 5 are caused by similar TTS voices, not by anything that necessarily happens with real responders. So “robust transcription and reliable semantic extraction across diverse operating conditions” is too strong for the evidence. The paper itself acknowledges this in Section VI, so it’s an overstatement in the abstract rather than a hidden flaw. Also, the GitHub link appears to contain only audio files, not the pipeline code or exact thresholds, which limits reproducibility.

None of this sinks the work. It’s a plausible feasibility study, and the central claim — that voice can serve as a semantic perception input for UAV networks — is reasonable. What’s missing is either a tempered abstract or, better, some real emergency/radio data to validate on. Given the scarcity of such data, the synthetic approach is defensible as a first pass.

Verdict: worth sending to peer review, not desk-rejecting. A good referee will ask for softened claims, error bars or more runs, and ideally one real-world sanity check. I’d bring it to a reading group if the group cares about LLM-plus-networking applications.

Referee Report

3 major / 5 minor

Summary. The paper introduces SIREN, a modular pipeline that combines automatic speech recognition (cloud AssemblyAI API or local Whisper) with LLM-based semantic extraction (Llama 3.2) and deterministic NLP validation (NER, speaker diarization, sentiment analysis) to convert emergency voice radio traffic into structured JSON outputs containing locations, responding units, emergency severity, and QoS expectations. The evaluation uses five synthetic emergency scenarios (English/Portuguese, clean/noisy, four or six TTS speakers) and reports transcription WER, output-quality success rates for location identification, unit extraction, speaker count estimation, and QoS expectation extraction, plus execution times. The results show that the API-based ASR degrades less under noise than local Whisper, and identify speaker diarization and geocoding ambiguity as the principal bottlenecks; the most complex scenario (Scenario 5) achieves 0% location identification. The paper concludes that SIREN demonstrates the feasibility of voice-driven semantic perception for UAV-assisted emergency networks and serves as a proof of concept for decision support.

Significance. If the central claims held, SIREN would be a timely contribution: it operationalizes a modular ASR+LLM+NLP pipeline for emergency voice traffic, produces schema-constrained JSON outputs, uses deterministic validation to reduce hallucination risk, and makes code and data publicly available. The head-to-head comparison of local Whisper against a cloud ASR API under controlled clean/noisy conditions is also useful for deployment choices. However, the current evidence supports feasibility on a synthetic corpus, not the stronger claim of 'robust transcription and reliable semantic extraction across diverse operating conditions' made in the abstract. The five-run-per-scenario statistics, the synthetic LLM/TTS data generation, and the failure of the hardest scenario on location extraction all need to be addressed before the paper's central claim can be accepted as stated.

major comments (3)

[Section IV-A and Abstract] The evaluation corpus is fully synthetic: dialogues are authored by LLMs, spoken by ElevenLabs TTS, and degraded with Python post-processing. The paper itself states in Section VI that validation with authentic emergency recordings is future work. Every success rate in Tables III and IV is therefore measured on this synthetic distribution, which lacks radio compression, channel artifacts, genuine speaker overlap, and spontaneous disfluencies. The abstract's claim of 'robust transcription and reliable semantic extraction across diverse operating conditions' and the conclusion's 'reliably extracting locations and responding units' go beyond the evidence; at most, the paper demonstrates feasibility on its synthetic distribution. There is also a partial circularity: the ground-truth dialogues are generated by LLMs while SIREN's extraction stage is LLM-based. The deterministic NER/diarization
[Section IV-D, Tables III and IV] All output-quality rates are proportions out of only five runs, so a single run changes a reported percentage by 20 points. No confidence intervals, per-run counts, or significance tests are reported. Moreover, the 'five independent runs' are repeated executions over the same audio files; they measure stochastic variability of the pipeline, not variability across emergency scenarios or speakers. The 0% cells (speaker count in Scenarios 2, 3, and 5; location in Scenario 5) and the 100% cells are therefore less robust than the table formatting suggests. The paper should report exact counts, confidence intervals or variability measures, and ideally more runs; otherwise the word 'reliable' in the abstract is not statistically justified.
[Section IV-D-2-e, Scenario 5; Section VI] The highest-complexity scenario achieves 0% location identification and 0% speaker count in both clean and noisy conditions, and only 60% unit extraction. This is the paper's own hardest operating condition, and it directly contradicts the unqualified statement in the abstract that SIREN demonstrates 'reliable semantic extraction across diverse operating conditions.' The Discussion and Conclusions do acknowledge these limitations, but the central claim should be qualified to reflect that two of the four metrics fail in the most complex scenario. A phrasing such as 'feasibility under controlled synthetic conditions, with location extraction and speaker attribution remaining open challenges' would be accurate.

minor comments (5)

[Section III-B] Typo: 'Asssembly API' should be 'AssemblyAI API' (also in Section IV-C).
[Fig. 2] The figure label 'Audios ASR' is unclear; consider 'Audio input' or 'Audio → ASR'.
[Section IV-C] The text says the mean over 10 iterations is reported for the noisy condition, but Tables I and II do not indicate which rows are averaged and over how many iterations. Please make the averaging explicit for both clean and noisy tables, or remove the ambiguity.
[Section IV-D-1] The four output-quality dimensions are assessed through 'manual inspection of the synthetic ground-truth dialogues.' Please specify the inspection protocol: how many annotators, whether they were blind to the expected output, and how disagreements were resolved. This is important because the metrics are the basis for the paper's semantic-extraction claims.
[Section IV-A] The dataset repository link contains a space ('SIREN-audio files') and appears to be a placeholder. Please check that the URL is valid and correctly formatted.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the synthetic-data issue is an external-validity limitation, not a circular derivation.

full rationale

The paper's claimed derivation chain is an empirical pipeline (audio -> ASR -> LLM-based extraction -> deterministic NLP validation -> structured output), and its evaluation metrics (WER, success rates) are measured against explicit ground-truth dialogues rather than derived from the system's own parameters. No equation or construction makes a predicted quantity identical to an input: the ground-truth dialogues are generated by LLMs and TTS, while extraction is performed by LLaMA 3.2 with schema constraints and SpaCy-based validation; these are distinct components and the system can fail (e.g., 0% speaker-count in Scenarios 2/3/5, 40% QoS in Scenario 1). The synthetic-data limitation is openly stated in Section IV-A and deferred to future work in Section VI ('validate the framework with authentic emergency recordings'), so it is a generalizability concern, not a self-definitional or fitted-input circularity. The self-citations ([7], [10]) appear only in related-work context and do not carry the central argument. Therefore no circular step can be exhibited under the required standard.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central claim rests mainly on the representativeness of synthetic emergency dialogues and on the reliable operation of external commercial APIs. The only hand-tuned numerical rule is the sentiment escalation threshold, which is not quantified. No new physical or conceptual entities are introduced.

free parameters (1)

Sentiment escalation threshold = not specified
In Section III-B.3, the emergency level is escalated when 'sentiment indicates a strong negative affect' and adjusted downward on neutral/positive sentiment; the numeric threshold on the mean sentiment score is hand-chosen but not reported, making the behavior non-replicable.

assumptions (4)

domain assumption Synthetic LLM/TTS dialogues are representative of real emergency voice traffic.
Section IV-A builds the entire evaluation on this assumption, with no real-world validation.
domain assumption Cloud ASR, diarization, sentiment, and geocoding APIs provide reliable outputs.
The implementation in Section III-B depends on AssemblyAI, SpaCy, and GeoPy behaving consistently; the paper even notes geocoding errors and diarization failures, so this assumption is partially violated.
domain assumption The fixed JSON schema is sufficient and appropriate for UAV network management decisions.
Section III-A.3 defines the output schema as linking to UAV positioning and resource allocation, but no actual network management algorithm consumes it in this paper.
domain assumption NER and sentiment validation improve LLM extraction without introducing systematic errors.
Section III-B describes these deterministic modules as reliability aids, but their individual error rates and interaction with LLM outputs are not evaluated separately.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Voice-Driven Semantic Perception for UAV-Assisted Emergency Networks." pith.science (2026). https://pith.science/paper/2P273I73

@misc{pith2026260217394,
  author       = {Pith},
  title        = {Pith review of: Voice-Driven Semantic Perception for UAV-Assisted Emergency Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2P273I73}},
  note         = {Machine review of arXiv:2602.17394}
}

read the original abstract

Unmanned Aerial Vehicle (UAV)-assisted networks are increasingly foreseen as a promising approach for emergency response, providing rapid, flexible, and resilient communications in environments where terrestrial infrastructure is degraded or unavailable. In such scenarios, voice radio communications remain essential for first responders due to their robustness; however, their unstructured nature prevents direct integration with automated UAV-assisted network management. This paper proposes SIREN, an AI-driven framework that enables voice-driven perception for UAV-assisted networks. By integrating Automatic Speech Recognition (ASR) with Large Language Model (LLM)-based semantic extraction and Natural Language Processing (NLP) validation, SIREN converts emergency voice traffic into structured, machine-readable information, including responding units, location references, emergency severity, and Quality-of-Service (QoS) requirements. SIREN is evaluated using synthetic emergency scenarios with controlled variations in language, speaker count, background noise, and message complexity. The results demonstrate robust transcription and reliable semantic extraction across diverse operating conditions, while highlighting speaker diarization and geographic ambiguity as the main limiting factors. These findings establish the feasibility of voice-driven situational awareness for UAV-assisted networks and show a practical foundation for human-in-the-loop decision support and adaptive network management in emergency response operations.

Figures

Figures reproduced from arXiv: 2602.17394 by the authors.

**Figure 2.** Overview of the SIREN pipeline, comprising ASR, Information Extraction, integrating probabilistic LLM inference with deterministic NLP validation, [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Interactive map interface depicting a geo-referenced unit and its [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Example of geocoding ambiguity in complex scenarios. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

20 extracted references · 1 linked inside Pith

[1]

Uavs meet llms: Overviews and perspectives towards agentic low-altitude mobility,

Y . Tian, F. Lin, Y . Li, T. Zhang, Q. Zhang, X. Fu, J. Huang, X. Dai, Y . Wang, C. Tian, B. Li, Y . Lv, L. Kov´acs, and F.-Y . Wang, “Uavs meet llms: Overviews and perspectives towards agentic low-altitude mobility,” Information Fusion, vol. 122, p. 103158, 2025. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S1566253525002313

2025
[2]

AirVista: Empowering UA Vs with 3D Spatial Reasoning Abilities Through a Multimodal Large Language Model Agent,

F. Lin, Y . Tian, Y . Wang, T. Zhang, X. Zhang, and F.-Y . Wang, “AirVista: Empowering UA Vs with 3D Spatial Reasoning Abilities Through a Multimodal Large Language Model Agent,” in2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), 2024, pp. 476–481

2024
[3]

Intelligent air traffic control using NLP-enhanced speech recognition and natural language generation,

A. Sarhan, R. Fathy, and H. Ali, “Intelligent air traffic control using NLP-enhanced speech recognition and natural language generation,” Journal of Electrical Systems and Information Technology, vol. 12, 07 2025

2025
[4]

Named entity recognition (NER) in NLP techniques, tools accuracy and performance,

S. Naseer, M. Ghafoor, S. Alvi, A. Kiran, G. Shafique Ur Rahmand, and G. Murtazae, “Named entity recognition (NER) in NLP techniques, tools accuracy and performance,”Pakistan Journal of Multidisciplinary Research, vol. 2, no. 2, pp. 293–308, Jan. 2022

2022
[5]

Utilizing UA Vs in Wireless Networks: Advantages, Challenges, Objectives, and Solution Methods,

M. J. Sobouti, A. Mohajerzadeh, H. Y . Adarbah, Z. Rahimi, and H. Ahmadi, “Utilizing UA Vs in Wireless Networks: Advantages, Challenges, Objectives, and Solution Methods,”Vehicles, vol. 6, no. 4, pp. 1769–1800, 2024. [Online]. Available: https://www.mdpi.com/2624- 8921/6/4/86

2024
[6]

A Survey on UA V-Assisted Wireless Communications: Challenges, Technologies, and Application,

S. A. Owaid, A. H. Miry, and T. M. Salman, “A Survey on UA V-Assisted Wireless Communications: Challenges, Technologies, and Application,” in2024 11th International Conference on Electrical and Electronics Engineering (ICEEE), 2024, pp. 333–340

2024
[7]

SUPPLY: Sustainable Multi- UA V Performance-Aware Placement Algorithm for Flying Networks,

P. Ribeiro, A. Coelho, and R. Campos, “SUPPLY: Sustainable Multi- UA V Performance-Aware Placement Algorithm for Flying Networks,” IEEE Access, vol. 12, pp. 159 445–159 461, 2024

2024
[8]

Vision-Based Learning for Drones: A Survey,

J. Xiao, R. Zhang, Y . Zhang, and M. Feroskhan, “Vision-Based Learning for Drones: A Survey,”IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 9, pp. 15 601–15 621, 2025

2025

Show all 20 references

[9]

From Sound to Sight: Audio-Visual Fusion and Deep Learning for Drone Detection,

I. Alla, H. B. Olou, V . Loscri, and M. Levorato, “From Sound to Sight: Audio-Visual Fusion and Deep Learning for Drone Detection,” inProceedings of the 17th ACM Conference on Security and Privacy in Wireless and Mobile Networks, ser. WiSec ’24, 2024, p. 123–133. [Online]. Ava...

2024
[10]

A4FN: an Agentic AI Architecture for Autonomous Flying Networks,

A. Coelho, P. Ribeiro, H. Fontes, and R. Campos, “A4FN: an Agentic AI Architecture for Autonomous Flying Networks,” in2025 IEEE 36th International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), 2025, pp. 1–6

2025
[11]

Whisper: OpenAI’s Speech Recognition Model,

OpenAI, “Whisper: OpenAI’s Speech Recognition Model,” https://github.com/openai/whisper, 2022, Accessed: Jan. 16, 2026

2022
[12]

AssemblyAI Documentation,

AssemblyAI, “AssemblyAI Documentation,” https://www.assemblyai.com/docs, 2025, Accessed: Jan. 16, 2026

2025
[13]

LLaMA 3: Open Foundation and Instruction Models,

H. Touvron, T. Lavril, G. Izacard, X. Martinetet al., “LLaMA 3: Open Foundation and Instruction Models,” 2024, meta AI. [Online]. Available: https://ai.meta.com/llama

2024
[14]

Ollama: Get up and running with large language models

Ollama, “Ollama: Get up and running with large language models.” https://ollama.com, 2024, Accessed: Jan. 16, 2026

2024
[15]

spaCy: Industrial-strength Natural Language Processing in Python,

M. Honnibal, I. Montani, S. Van Landeghem, and A. Boyd, “spaCy: Industrial-strength Natural Language Processing in Python,” 2020. [Online]. Available: https://doi.org/10.5281/zenodo.1212303

2020 doi
[16]

GeoPy: Geocoding library for Python,

GeoPy, “GeoPy: Geocoding library for Python,” Accessed: 2026-1-16. [Online]. Available: https://geopy.readthedocs.io/en/stable/

2026
[17]

16, 2026

Folium, “Folium,” Accessed: Jan. 16, 2026. [Online]. Available: https://python-visualization.github.io/folium/latest/

2026
[18]

RescueSpeech: A German Corpus for Speech Recognition in Search and Rescue Domain,

S. Sagar, M. Ravanelli, B. Kiefer, I. K. Korbayova, and J. van Genabith, “RescueSpeech: A German Corpus for Speech Recognition in Search and Rescue Domain,” 2023. [Online]. Available: https://arxiv.org/abs/2306.04054

2023 arXiv
[19]

Text to Speech — ElevenLabs Documenta- tion,

ElevenLabs, “Text to Speech — ElevenLabs Documenta- tion,” 2025, Accessed: Jan. 16, 2026. [Online]. Available: https://elevenlabs.io/docs/capabilities/text-to-speech

2025
[20]

Minimizing Word Error Rate in Tex- tual Summaries of Spoken Language,

K. Zechner and A. Waibel, “Minimizing Word Error Rate in Tex- tual Summaries of Spoken Language,”Language Technologies Insti- tute, Carnegie Mellon University, 2000, https://aclanthology.org/A00- 2025.pdf

2000

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.

[1] [1]

Uavs meet llms: Overviews and perspectives towards agentic low-altitude mobility,

Y . Tian, F. Lin, Y . Li, T. Zhang, Q. Zhang, X. Fu, J. Huang, X. Dai, Y . Wang, C. Tian, B. Li, Y . Lv, L. Kov´acs, and F.-Y . Wang, “Uavs meet llms: Overviews and perspectives towards agentic low-altitude mobility,” Information Fusion, vol. 122, p. 103158, 2025. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S1566253525002313

2025

[2] [2]

AirVista: Empowering UA Vs with 3D Spatial Reasoning Abilities Through a Multimodal Large Language Model Agent,

F. Lin, Y . Tian, Y . Wang, T. Zhang, X. Zhang, and F.-Y . Wang, “AirVista: Empowering UA Vs with 3D Spatial Reasoning Abilities Through a Multimodal Large Language Model Agent,” in2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), 2024, pp. 476–481

2024

[3] [3]

Intelligent air traffic control using NLP-enhanced speech recognition and natural language generation,

A. Sarhan, R. Fathy, and H. Ali, “Intelligent air traffic control using NLP-enhanced speech recognition and natural language generation,” Journal of Electrical Systems and Information Technology, vol. 12, 07 2025

2025

[4] [4]

Named entity recognition (NER) in NLP techniques, tools accuracy and performance,

S. Naseer, M. Ghafoor, S. Alvi, A. Kiran, G. Shafique Ur Rahmand, and G. Murtazae, “Named entity recognition (NER) in NLP techniques, tools accuracy and performance,”Pakistan Journal of Multidisciplinary Research, vol. 2, no. 2, pp. 293–308, Jan. 2022

2022

[5] [5]

Utilizing UA Vs in Wireless Networks: Advantages, Challenges, Objectives, and Solution Methods,

M. J. Sobouti, A. Mohajerzadeh, H. Y . Adarbah, Z. Rahimi, and H. Ahmadi, “Utilizing UA Vs in Wireless Networks: Advantages, Challenges, Objectives, and Solution Methods,”Vehicles, vol. 6, no. 4, pp. 1769–1800, 2024. [Online]. Available: https://www.mdpi.com/2624- 8921/6/4/86

2024

[6] [6]

A Survey on UA V-Assisted Wireless Communications: Challenges, Technologies, and Application,

S. A. Owaid, A. H. Miry, and T. M. Salman, “A Survey on UA V-Assisted Wireless Communications: Challenges, Technologies, and Application,” in2024 11th International Conference on Electrical and Electronics Engineering (ICEEE), 2024, pp. 333–340

2024

[7] [7]

SUPPLY: Sustainable Multi- UA V Performance-Aware Placement Algorithm for Flying Networks,

P. Ribeiro, A. Coelho, and R. Campos, “SUPPLY: Sustainable Multi- UA V Performance-Aware Placement Algorithm for Flying Networks,” IEEE Access, vol. 12, pp. 159 445–159 461, 2024

2024

[8] [8]

Vision-Based Learning for Drones: A Survey,

J. Xiao, R. Zhang, Y . Zhang, and M. Feroskhan, “Vision-Based Learning for Drones: A Survey,”IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 9, pp. 15 601–15 621, 2025

2025