Systematic Literature Review

Template Overview

A rigorous systematic literature review demands transparent search methodologies, explicit inclusion and exclusion criteria, and reproducible data extraction workflows. Following the PRISMA reporting statement, this Markdown template structures academic database queries, deduplication records, bias assessments, and thematic syntheses into an auditable research dossier suitable for peer review submission.

Markdown Template Structure

Copy the block below or click Download .md File to add this template directly to your personal vault:

---
title: "Systematic Review: [Research Domain / Question]"
date: [YYYY-MM-DD]
primary_investigator: "[[Lead Researcher]]"
co_investigators: ["[[Co-Author 1]]", "[[Co-Author 2]]"]
prisma_registration: "[PROSPERO ID / OSF Link]"
databases_searched: ["PubMed", "IEEE Xplore", "ACM Digital Library", "Web of Science"]
tags: [academic-research, systematic-review, prisma, literature-review, methodology]
---

# Systematic Literature Review — [Research Domain]

## 1. PICO Framework & Research Question
- **Population (P):** `[Target demographic, system, or dataset]`
- **Intervention (I):** `[Core technique, algorithm, or methodology being evaluated]`
- **Comparison (C):** `[Baseline, benchmark, or control method]`
- **Outcome (O):** `[Primary quantitative or qualitative performance metrics]`
- **Primary Research Question:** How does [Intervention] compare against [Comparison] in improving [Outcome] within [Population]?

## 2. Search Strategy & Database Query Syntax
- **PubMed Syntax:** `("on-device machine learning" OR "local inference") AND ("speech-to-text" OR "automatic speech recognition") AND ("privacy" OR "confidentiality")`
- **IEEE Xplore Syntax:** `(("Document Title":on-device) AND ("Abstract":ASR) AND ("Abstract":privacy))`
- **Date Range Parameters:** 2018-01-01 to 2026-10-01
- **Language Restrictions:** English peer-reviewed conference proceedings and journals only.

## 3. Inclusion & Exclusion Criteria
| Criterion | Inclusion Rules | Exclusion Rules |
|---|---|---|
| Study Type | Randomized trials, empirical benchmarks, controlled user evaluations | Opinion pieces, non-peer-reviewed blog posts, pre-patent filings |
| Architecture | On-device, edge computing, Apple Silicon / NPU deployments | Server-only cloud inference architectures with no local fallback |
| Language Support | Multi-lingual or English speech corpus validation | Synthesized text-to-speech studies without ASR recognition |

## 4. PRISMA Screening Ledger
- **Records Identified Across Databases:** [N = 1,420]
- **Duplicate Records Removed:** [N = 380]
- **Records Screened by Title & Abstract:** [N = 1,040]
- **Full-Text Articles Assessed for Eligibility:** [N = 112]
- **Studies Included in Qualitative Synthesis:** [N = 34]

## 5. Synthesis & Risk of Bias Assessment
| Paper Citation | Methodological Approach | Sample Size / Benchmark | Risk of Bias (ROBINS-I) | Core Contribution |
|---|---|---|---|---|
| Chen et al. (2024) | Quantized Transformer on Core ML | LibriSpeech test-clean (N=5.4h) | Low Risk | 3.4x throughput speedup with <0.2% WER degradation |
| Vance & Thorne (2025) | Local Whisper on Edge NPUs | Real-world clinical audio (N=48h) | Low Risk | Complete air-gapped compliance with zero cloud leakage |

## 6. Action Items & Manuscript Preparation
- [ ] Draft PRISMA flow diagram visualization.
- [ ] Complete cross-extraction data validation with second reviewer.
- [ ] Export BibTeX citations to project reference manager.

Filled Real-World Example

Here is a realistic example demonstrating how this template captures actual deliberations:

---
title: "Systematic Review: On-Device Neural Speech Recognition"
date: 2026-10-09
primary_investigator: "[[Dr. Alistair Finch]]"
co_investigators: ["[[Elena Rostova]]", "[[Marcus Vance]]"]
prisma_registration: "OSF.IO/X879A"
databases_searched: ["IEEE Xplore", "ACM DL", "arXiv cs.CL", "PubMed"]
tags: [academic-research, systematic-review, prisma, literature-review, methodology]
---

# Systematic Literature Review — On-Device Neural Speech Recognition

## 1. PICO Framework & Research Question
- **Population:** Knowledge workers, medical practitioners, and legal professionals requiring real-time speech transcription.
- **Intervention:** Quantized on-device transformer models (e.g., Whisper Core ML, GGML Whisper) running locally on consumer hardware.
- **Comparison:** Cloud-hosted ASR APIs (OpenAI Whisper API, Google Speech-to-Text, AWS Transcribe).
- **Outcome:** Word Error Rate (WER), real-time factor latency (RTF), memory footprint, and data exfiltration risk.
- **Research Question:** Can local edge-quantized speech-to-text match cloud API transcription accuracy while guaranteeing zero network transmission?

## 2. Inclusion & Exclusion Criteria
- **Included:** Peer-reviewed papers measuring quantized ASR models on Apple Silicon or mobile NPUs from 2022 to 2026.
- **Excluded:** Studies lacking reproducible open-source benchmarks or requiring server-side assistance.

## 3. PRISMA Screening Summary
- Initial Database Hits: 842 records.
- Deduplicated: 615 unique titles.
- Full-Text Screened: 64 candidate papers.
- Final Included Synthesis: 21 rigorous benchmark studies.

## 4. Key Findings & Synthesis
- 4-bit and 5-bit quantized models running on unified memory architectures achieve statistical parity with float32 cloud endpoints on clean speech (mean WER delta < 0.4%).
- Thermal throttling during sustained hour-long audio sessions is mitigated through 8-core neural engine scheduling.
- Local air-gapped inference completely eliminates third-party subpoena vulnerability and cloud data interception risks.

## 5. Action Items
- [ ] Generate forest plots comparing WER across model parameter sizes.
- [ ] Format methodology section for IEEE Transactions on Audio, Speech, and Language Processing.
- [ ] Sync final citations with Zotero collection.

When to Use and When to Avoid

When to Use

• Use when executing formal systematic reviews, scoping reviews, or meta-analyses for academic publication.
• Use to establish transparent audit trails for medical, legal, or technical literature syntheses.
• Use during doctoral dissertation literature mapping to prevent selection bias.

When to Avoid

• Avoid using for casual weekly reading lists or quick informal blog post research.
• Avoid using when evaluating single papers in isolation (use research paper critique instead).

On-Device AI Integration

Summarize your recording in Oyma with this local instruction:

Analyze the systematic literature review notes:
1. Validate search query strings against standard Boolean query logic for PubMed and IEEE Xplore.
2. Verify PRISMA screening counts for mathematical consistency across initial hits, exclusions, and included studies.
3. Draft a three-paragraph narrative synthesis summarizing the primary methodological findings.
Output in clean Markdown.

Your data remains securely on your Mac with zero cloud exposure.