Singing Detection via MIDI Note Scoring on Low-Power MCUs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing singing detection technologies based on deep learning models require significant computational resources and energy, making them unsuitable for low-power MCUs, which limits their application in embedded systems and IoT devices.
Innovation Solution
A singing detection method and system that utilizes packet trigger judgment, tempo detection, pitch detection, and quantization to convert audio signals into MIDI note data, performing weighted analysis of rhythm and pitch variations without requiring high-performance processors or large memory.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning models are used for singing detection, then detection accuracy and robustness are improved, but computational resource requirements and energy consumption increase significantly
Solution Approach 1:
The patent divides the singing detection task into multiple independent modules: packet trigger judgment, tempo detection, pitch detection, quantization operation, and rhythm/pitch variation judgment. Each module processes specific features separately, allowing the system to achieve accurate detection while reducing overall computational load by only processing relevant features rather than using a monolithic deep learning model.
Solution Approach 2:
The patent extracts and processes only the most critical features for singing detection (tempo, pitch, rhythm variations) rather than using comprehensive deep learning models that process all audio characteristics. This selective feature extraction approach maintains detection accuracy while significantly reducing computational resource requirements and energy consumption.
2Measurement precision
If deep learning models are used for singing detection, then detection accuracy and robustness are improved, but hardware resource requirements increase
Solution Approach 1:
The patent segments the detection system into multiple simple, independent processing modules that can be implemented on low-power MCUs. Each module performs a specific function (packet judgment, tempo detection, pitch detection) using basic signal processing operations rather than complex neural network layers, reducing hardware requirements while maintaining accuracy.
Solution Approach 2:
The patent uses simple, lightweight algorithms and data structures that can be easily implemented on resource-constrained devices. Instead of requiring powerful GPUs or FPGAs, the system uses basic arithmetic operations, lookup tables, and simple filters that can run on inexpensive microcontrollers, making the technology accessible for embedded applications.
3Measurement precision
If complex neural network models are used for singing detection, then detection performance is improved, but processing time increases
Solution Approach 1:
The patent segments the audio processing into discrete time-based operations (packet trigger judgment, tempo detection, pitch detection) that can be executed sequentially or in parallel on simple hardware. This modular approach allows for optimized processing timing and reduces the computational time required compared to executing complex neural network inference on the same hardware.
Solution Approach 2:
The patent applies partial processing by only analyzing audio packets that meet specific criteria (packet trigger judgment) and only processing features that are most indicative of singing (tempo, pitch variations). This selective processing approach reduces the total computation time while maintaining sufficient detection performance without the overhead of processing all possible audio characteristics.
Data Source
AI summary
A singing detection method and a singing detection system using the same are provided in the embodiments of the present invention. The singing detection method includes the following steps: performing packet trigger judgment to activate the detection function; conducting rhythm detection on the sound signal; performing pitch detection on the sound signal; quantizing the pitch detection data; converting the quantized data and rhythm detection data into MIDI note data; using the MIDI note data to judge rhythm and pitch changes; assigning weighted scores and comparing with a threshold score. Finally, it determines whether the sound signal is singing based on the total score. This method combines multiple musical feature analyses to identify singing voices through comprehensive scoring.