S3
ESP32-S3 Micro-Inference Profiler
Audio Model Allocator & Latency Inspector
SRAM: 512 KB Internal
PSRAM: 8 MB Octal SPI
ESP-DSP Vector Accelerators
Model Architecture
Presets Available
Architecture Preset
Micro-Keyword Spotter (250K)
Local Intent Classifier (1.0M)
Minimal Embedding Extractor (500K)
Custom Parameters
Model Parameters
250,000
Quantization Precision
INT4 (0.5 B/param - Sub-byte Pack)
INT8 (1.0 B/param - Quantized TFLite)
FP16 (2.0 B/param - Half Precision)
Audio Frontend & DSP
Sample Rate & Feature
16 kHz - MFCC (40 coeffs)
16 kHz - MFE Spectrogram (64 bins)
8 kHz - MFCC (26 coeffs)
Audio Buffer Duration
1000 ms
Memory Allocation Strategy
Auto-Staging (Weights PSRAM / TensorArena SRAM)
Strict SRAM Only (Zero PSRAM Latency)
PSRAM Heavy Arena (SRAM Stack Conservation)
Simulate Profile
Reset Defaults
✔ Memory layout valid. Arena fits within internal SRAM constraints.
SRAM Usage
142 KB / 512 KB
PSRAM Usage
244 KB / 8192 KB
Inference Latency
18.4 ms
CPU Cycles / Frame
4.41 M cycles
SRAM vs PSRAM Physical Allocation Map
ESP32-S3 Dual-Core LX7 @ 240MHz
Local Vector Keyword Distance Simulator (ESP-DSP Vector Math)
Simulate real-time cosine distance between live MFCC buffer vector and stored wake-word vector.
Voice Similarity:
0.82
Match Threshold:
0.75
[ WAKE WORD DETECTED ]
ESP-IDF C++ Header
Copy Code
// Generating code...
Download esp32_audio_config.h
Refocus Primary Controls
Enjoy this tool? Build your own with Super