A unique feature of v1 is the --mute flag, which strips all high-intensity sentiment from input text, rendering it neutral. This is designed for content moderation systems that need to de-escalate aggressive communication automatically.
| Metric | Value | |--------|-------| | Accuracy (on RAF-DB + IEMOCAP) | 84.3% | | Latency (full pipeline) | 45ms (M1 CPU) / 22ms (NVIDIA T4) | | Model Size | 14.8 MB | | RAM Usage | ~180 MB | | Power Draw | 1.2W (inference on Raspberry Pi 4) | emous v1
is the inaugural release of a lightweight, high-speed sentiment analysis engine designed for embedded systems and low-latency applications. Where traditional NLP models focus on complex contextual understanding, emous focuses on the raw, atomic weight of language—stripping away the syntax to leave only the feeling. A unique feature of v1 is the --mute
: Classic versions of Paint , Winamp , and even the helpful (or persistent) Clippy . Where traditional NLP models focus on complex contextual
“Emous V1 doesn’t pretend to read minds – it reads faces, voices, and words, then responds appropriately. And that’s enough for most real‑world products.”
| V1 (Current) | V2 (Planned) | |--------------|----------------| | Single emotion | Mixed emotion vectors | | No temporal memory | LSTM-based emotion history | | 7 emotions | 12 emotions (incl. contempt, interest, confusion) | | CPU/GPU only | NPU + microcontroller support | | English only | 5 languages (en, es, zh, ar, hi) |