Name: silero vad coreml
Author: FluidInference

Overview

Silero VAD CoreML is a specialized voice activity detection model optimized for Apple's hardware acceleration. Unlike general-purpose audio models, this implementation focuses on high-precision speech boundary detection with minimal latency, making it ideal for real-time streaming applications. For developers, the CoreML conversion means significantly lower CPU overhead and better battery efficiency on iOS and macOS compared to standard PyTorch versions. It serves as an efficient front-end filter for STT (Speech-to-Text) pipelines, ensuring that heavy transcription models are only triggered when actual speech is detected. Integration is straightforward for those already within the Apple ecosystem, offering a lightweight alternative to cloud-based VAD services without sacrificing accuracy.

Highlights

Hardware-accelerated inference for iOS and macOS devices
Low-latency speech boundary detection for real-time apps
Reduces compute costs by filtering non-speech audio
MIT licensed for flexible commercial integration
Optimized energy efficiency via CoreML framework

Usage

Install

# Install Hugging Face transformers
pip install transformers torch

SDK Usage

# Load model with transformers
from transformers import AutoModel, AutoTokenizer

model = AutoModel.from_pretrained("FluidInference/silero-vad-coreml")
tokenizer = AutoTokenizer.from_pretrained("FluidInference/silero-vad-coreml")

Hugging Face Download

We recommend downloading the model via the Hugging Face CLI or Hub SDK.

Guidance：Before downloading, install huggingface_hub with:

Guidance

pip install -U huggingface_hub

CLI Download

Download the full repository

huggingface-cli download FluidInference/silero-vad-coreml

Download a single file to a local folder (e.g. config.json into ./dir)

huggingface-cli download FluidInference/silero-vad-coreml config.json --local-dir ./dir

See the official docs for more CLI options

SDK Download

# 模型下载
from huggingface_hub import snapshot_download
model_dir = snapshot_download('FluidInference/silero-vad-coreml')

Git Download

Make sure git-lfs is installed first

Git Download

git lfs install
git clone https://huggingface.co/FluidInference/silero-vad-coreml

To skip LFS large-file downloads, use:

Skip LFS

GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/FluidInference/silero-vad-coreml

Model files are hosted on the Hugging Face Hub — download directly via HF CLI / SDK / Git, not through this site.

PyTorch / Transformers Usage

Install Transformers

pip install -U transformers torch

Load the model and run inference

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained('FluidInference/silero-vad-coreml')
tokenizer = AutoTokenizer.from_pretrained('FluidInference/silero-vad-coreml')

Full Documentation

来源: HuggingFace

---
license: mit
tags:

audio

voice-activity-detection

coreml

silero

speech

ios

macos

swift

library_name: coreml
pipeline_tag: voice-activity-detection
datasets:

alexwengg/musan_mini50

alexwengg/musan_mini100

metrics:

accuracy

f1

language:

en

base_model:

onnx-community/silero-vad

---

<span style="color:#5DAF8D">🧃 CoreML Silero VAD </span>

![Discord](https://discord.gg/WNsvaCtmDe) ![GitHub Repo stars](https://github.com/FluidInference/FluidAudio)

A CoreML implementation of the Silero Voice Activity
Detection (VAD) model, optimized for Apple platforms
(iOS/macOS). This repository contains pre-converted
CoreML models ready for use in Swift applications.

See FluidAudio Repo link at the top for more information

Model Description

Developed by: Silero Team (original), converted by
FluidAudio

Model type: Voice Activity Detection

License: MIT

Parent Model:
silero-vad

This is how the model performs against the silero-vad v6.0.0 basline Pytorch JIT version

!graphs/yc_standard_comparison_20250915_205721_2c04b81.png
!graphs/yc_256ms_comparison_20250915_205721_2c04b81.png

Note that we tested the quantized versions, as the model is already tiny, theres no performance imporvement at all.

This is how the different models compare in terms of speed, the 256s takes in 8 chunks of 32ms and processes it in batches so its much faster
!graphs/yc_performance_20250915_205721_2c04b81.png

Conversion code is available here: FluidInference/mobius

Intended Use

Primary Use Cases

Real-time voice activity detection in iOS/macOS

applications

Speech preprocessing for ASR systems

Audio segmentation and filtering

How to Use

Citation

@misc{silero-vad-coreml,
title={CoreML Silero VAD},
author={FluidAudio Team},
year={2024},

url={https://huggingface.co/alexwengg/coreml-silero-vad}
}

@misc{silero-vad,
title={Silero VAD},
author={Silero Team},
year={2021},
url={https://github.com/snakers4/silero-vad}
}

GitHub: https://github.com/FluidAudio/FluidAudioSwift