FireRedVAD-GGUF

FireRedVAD GGUF Models

Pre-converted FireRedVAD models in GGUF format for cross-platform, high-performance C++ inference. Includes Standard VAD, Streaming VAD, and Audio Event Detection (AED) with multiple quantization options.

• Original Models: FireRedTeam/FireRedVAD
• C++ Inference Engine & Conversion Tools: https://github.com/Strg-Alt-Entf-0x00/firered-vad

(Update 2026-08-30: Models re-uploaded to fix a mathematical transposition bug in the FSMN lookahead filters. These models now have 100% exact parity with the original PyTorch implementation).

⚡ Quick Start (Python / Hub)

pip install huggingface-hub
from huggingface_hub import hf_hub_download

# Download streaming VAD (INT8-CH, recommended)
model_path = hf_hub_download(
    repo_id="Strg-Alt-Entf-0x00/FireRedVAD-GGUF",
    filename="stream-vad/int8-ch/firered-stream-vad-int8-ch.gguf"
)

For the highly-optimized Native C++ Engine (no PyTorch required, up to 53x realtime), see the Integration Repository.

---

📁 Available Models

All models follow this structure: {model_type}/{quantization}/{filename}.gguf

stream-vad/ — Real-time Streaming VAD (Causal, Low Latency)

Best for real-time voice activity detection. Fully causal — no future context needed. Optimized for streaming applications with 10ms frame processing. Achieves up to 53.6x realtime performance on a single CPU core.

vad/ — Standard VAD (Bidirectional, High Accuracy)

Uses bidirectional context (lookback + lookahead). Higher accuracy than stream-vad, but requires buffering full audio segments. Best for offline/batch processing. Performance: 97.57% F1 on FLEURS-VAD-102 dataset. Runs at 31.3x realtime.

aed/ — Audio Event Detection (Multi-class)

Simultaneous detection of speech, music, and singing in 100+ languages. Runs at 31.7x realtime.

---

🔍 Quantization Details

Why INT8-CH (Per-Channel) is Recommended

Problem with standard INT8 (per-tensor): DFSMN architectures have wide variance in weight distribution across output channels. A single global scale factor per tensor cannot capture this range accurately, causing silent accuracy degradation.

Solution: INT8 Per-Channel (int8-ch): Assigns one scale factor per output channel. This preserves near-FP32 accuracy (MAE < 0.001, SQNR > 59 dB) while maintaining INT8 speed and memory efficiency.

Quantization Quality Metrics (Per Model)

Each model includes a -debug.json file with per-tensor quantization statistics:

• MAE (Mean Absolute Error): Average difference from FP32
• SQNR (Signal-to-Quantization-Noise Ratio): Quality metric in dB
• Per-channel scale factors: Min/max/mean scales per layer

---

⚙️ Hardware Requirements

Minimum (INT8-CH models)

• RAM: ~5 MB for model + ~2 MB for inference buffers
• CPU: Any modern x86_64, ARM, or RISC-V processor
• Storage: ~600 KB per model

Recommended (FP32 models)

• RAM: ~10 MB for model + ~3 MB for inference buffers
• CPU: SIMD support (SSE, AVX, NEON) for faster inference
• Storage: ~2.3 MB per model

📜 License & Attribution

• FireRedVAD by Xiaohongshu (FireRedTeam) — Apache 2.0
  Source: https://github.com/FireRedTeam/FireRedVAD
• FireRedVAD-GGUF by Strg-Alt-Entf-0x00 — Apache 2.0
  Repository: https://github.com/Strg-Alt-Entf-0x00/firered-vad