top of page
圖片1.jpg
圖片4.jpg

​最後更新2025/11/18

全方位 AI 智慧長照：從精準跌倒偵測到即時語音陪伴

 

 

喬泰科技致力於打造有溫度的科技照護。我們基於 Intel Core Ultra 7 處理器 的高效能邊緣運算平台，不僅具備鷹眼般的視覺偵測能力，更整合了先進的語音識別 (ASR) 與大型語言模型 (LLM) 技術，為長照場域提供「視、聽、意」三位一體的智慧守護方案。

Google_AI_Studio_2025-11-18T12_01_50.762Z.png

1. 視覺守護：即時姿態分析與跌倒偵測

 

系統核心採用 YOLOv8/YOLO11 Pose 姿態估計模型，能精準捕捉人體骨骼關鍵點，即時判斷長者是否發生跌倒或長時間無動作等異常狀況 。經實測，在 Intel 平台上運行時，延遲變化的標準差控制在 5.11ms 以內 ，確保在危急時刻能穩定、零時差地發出警示。

output_fall_down.gif
image.png

  • 不同模型大小效能、性能比較：YOLO11 模型架構 - pose 任務

image.png

  • 模型參數量對平均單幀模型推理延遲 Latency(ms) 的效能測試結果

2. 聽覺感知：高冗餘度的語音識別 (ASR)

 

除了看見危險，系統也能聽見呼救。我們整合了 OpenVINO™ 優化的 Whisper 語音模型，實現高精度的語音轉文字功能。

 

​效能實證：根據我們的測試報告，系統在處理語音轉譯時展現了極高的效能冗餘量 (Performance Redundancy) 。人類說話頻率約每秒 6 字，而本設備的運算能力遠超此需求 ，這意味著系統可以在進行高負載影像分析的同時，同步精準地捕捉長者的語音指令或是「救命」、「幫忙」等關鍵字，觸發緊急通報。

image.png

3. 智慧大腦：本地端大型語言模型 (LLM) 互動

 

為了給予長者更人性化的陪伴，我們在邊緣裝置上成功部署了 Llama-3.2 與 Gemma-2 等大型語言模型 。

​

流暢對話體驗：透過 INT4 量化技術優化，即便是參數高達 9B 的模型，在我們的 Intel 單板機上也能達到 每秒生成 7.11 至 17.28 個 token 的速度 。這項數據證實了模型的生成速度已等同甚至超越了人類的說話速度 ，能夠實現自然、無延遲的語音互動。

隱私與陪伴：所有運算皆在本地端完成，長者的對話內容無需上傳雲端，絕對保障隱私。系統可扮演貼心的照護助理，不僅能回覆日常問題，還能安撫長者情緒，實現真正的智慧陪伴。

image.png

  • 大型語言模型 Large Language Model 效能測試結果

喬泰科技整合 Intel Core Ultra 的 CPU、GPU 與 NPU 算力，將「跌倒偵測」、「語音求助」與「智慧對話」完美融合於單一設備。我們不只提供安全的監控，更提供即時、懂你、且無所不在的 AI 照護服務。

S__73326627.jpg

測試環境：

 

  • CPU核心數：16核心

  • Product system: Gigabyte brix

  • Model name：Intel(R) Core(TM) Ultra 7 255H

  • 總容量：64GB （32GBx2）DDR5

  • 作業系統：Ubuntu 24.04.3 LTS

  • Python 版本：3.12.3

  • OpenVINO 版本：2025.1.0​

GB-BRU7-255H 是一款高效能的 AI 迷你電腦，賦予本公司 AI 系統更強大的落地能力。無論是需要高度隱私與即時反應的「長期照護監控」，還是追求高互動性的「AI智能客服」，255H 都能憑藉 Intel Core Ultra 的 AI 算力輕鬆駕馭。小巧機身隱身於任何角落，卻能提供工作站等級的運算效能，是我們實現智慧轉型、提升服務品質的最佳硬體夥伴。

作為 Intel Partner Alliance 生態系的一員，我們持續獲得英特爾技術團隊的專業支援，涵蓋產品開發優化、市場協作及技術培訓，確保我們的服務緊扣最新技術趨勢。無論是雲端運算、AI 人工智慧、邊緣應用或物聯網領域，雙方合作將加速實現客戶的數位轉型目標。

 

未來，喬泰科技將秉持創新驅動的核心理念，攜手英特爾推動技術邊界，為產業創造更高價值。

​

探索技術可能性，即刻與我們聯繫

Comprehensive AI Smart Elderly Care: From Precise Fall Detection to Real-Time Voice Companionship

 

 

JTBC is dedicated to creating technology with warmth. Based on a high-performance edge computing platform powered by the Intel Core Ultra 7 processor , we integrate keen visual detection capabilities with advanced Automatic Speech Recognition (ASR) and Large Language Model (LLM) technologies. This delivers a "Vision, Hearing, and Intent" tri-fold smart guardianship solution for long-term care environments.

Google_AI_Studio_2025-11-18T12_01_50.762Z.png

1. Visual Guardianship: Real-Time Pose Analysis & Fall Detection

 

The core of our system utilizes advanced YOLOv8 and YOLO11 Pose estimation models, capable of precisely capturing human skeletal keypoints to instantly identify abnormalities such as falls or prolonged inactivity. Field tests confirm that when running on this Intel platform, the standard deviation of latency variation is controlled within 5.11ms. This ensures that in critical moments, alerts are issued stably and with zero delay.

output_fall_down.gif
image.png

  • Performance Comparison of Different Model Sizes: YOLO11 Model Architecture - Pose Task

image.png

  • Performance test results of model parameter quantity on average single-frame model inference latency (ms).

2. Auditory Perception: High-Redundancy Automatic Speech Recognition (ASR)

 

Beyond seeing danger, the system can hear calls for help. We have integrated OpenVINO™-optimized Whisper voice models to achieve high-precision speech-to-text functionality.

 

Performance Verification: Our test reports demonstrate significant performance redundancy. While human conversation frequency is typically under 6 words per second, our device's computing capability far exceeds this demand. This means the system can perform high-load image analysis while simultaneously and precisely capturing keywords like "Help" or voice commands, triggering emergency notifications without bottlenecks.

image.png

3. Smart Brain: Local Large Language Model (LLM) Interaction

 

To provide more human-centric companionship, we have successfully deployed Large Language Models like Llama-3.2 and Gemma-2 on our edge devices.

​

Fluid Conversation Experience: Through INT4 quantization optimization, models achieve generation speeds of up to 17.28 tokens per second (e.g., Llama-3.2-1b) on our Intel single-board computer. This speed matches or even surpasses human speech speed, enabling natural, latency-free voice interaction.

 

Privacy & Companionship: All computations are performed locally on the edge, ensuring that private conversations are never uploaded to the cloud. The system acts as a thoughtful care assistant, capable of answering daily queries and providing emotional support.

image.png

  • Performance test results of large language model

JTBC integrates the CPU, GPU, and NPU computing power of the Intel Core Ultra, seamlessly fusing "Fall Detection," "Voice Assistance," and "Smart Dialogue" into a single device. We provide not just secure monitoring, but an immediate, understanding, and omnipresent AI care service.

S__73326627.jpg

Test Environment:

 

  • CPU Core Count: 16 Cores

  • Product system: Gigabyte brix

  • Model Name: Intel(R) Core(TM) Ultra 7 255H

  • Total Memory: 64GB (32GBx2) DDR5

  • Operating System: Ubuntu 24.04.3 LTS

  • Python Version: 3.12.3

  • OpenVINO Version: 2025.1.0

The GB-BRU7-255H is a high-performance AI mini PC designed to empower the deployment of our proprietary AI systems.Whether for "Long-term Care Monitoring," which demands strict privacy and real-time responsiveness, or "AI Smart Customer Service" requiring high interactivity, this device effortlessly handles the workload using the advanced AI computing power of the Intel® Core™ Ultra processor. Its ultra-compact chassis fits discreetly into any environment while delivering workstation-grade performance, making it the ideal hardware partner for achieving smart transformation and elevating service quality.

Partnering with Intel: Co-Creating the Future of AI and IoT

​

As a proud member of the Intel Partner Alliance, JTBC receives ongoing support from Intel’s technical team—including product optimization, joint go-to-market efforts, and advanced training. This ensures our services are aligned with the latest technology trends across cloud computing, artificial intelligence, edge applications, and the Internet of Things.

Looking ahead, JTBC remains committed to innovation-driven growth. Together with Intel, we will continue pushing the boundaries of what’s possible—delivering transformative value to industries worldwide.

 

Discover the Possibilities of Innovation – Contact Us Today

 
bottom of page