Alkd/Silero-VAD-v5-CoreML

Silero-VAD-v5 — CoreML

CoreML conversion of Silero VAD v5 for Apple Neural Engine.

Model Details

Detail	Value
Architecture	STFT → Conv1d encoder → LSTM → decoder
Parameters	~309K
Input	512 samples (32ms @ 16kHz)
Output	Speech probability (0.0–1.0)
Size	~4.2 MB

Usage

let vad = try await SileroVADModel.fromPretrained(backend: .coreML)
let prob = vad.processChunk(samples)

Variants

Variant	Backend	Model ID
MLX	GPU	aufklarer/Silero-VAD-v5-MLX
CoreML	Neural Engine	aufklarer/Silero-VAD-v5-CoreML

Links

• Swift library: soniqo/speech-swift
• Original model: snakers4/silero-vad

---

Links

• Blog: blog.ivan.digital
• Library Docs: soniqo.audio