FireRed VAD Stream

这是由原始 FireRedTeam/FireRedVAD/Stream-VAD 转换得到的标准 Hugging Face 模型仓库。

仓库内容：

model.safetensors：流式 VAD 模型权重
config.json：firered-vad-stream 模型配置
cmvn.ark：与原始模型一致的 CMVN 统计量

该仓库不包含 HF remote code。模型结构定义在 fireredvad 项目内：

fireredvad.models.firered_vad_stream.configuration_firered_vad_stream.FireRedVadStreamConfig
fireredvad.models.firered_vad_stream.modeling_firered_vad_stream.FireRedVadStreamModel
fireredvad.models.firered_vad_stream.module_firered_vad_stream.FireRedVadStreamModule

服务加载示例：

poetry run python -m fireredvad.commands.app MigoXV/firered-vad-stream \
  --device cpu \
  --host 0.0.0.0 \
  --port 50051 \
  --stream-chunk-frame 10 \
  --no-cuda-graph

直接加载示例：

from fireredvad.models.firered_vad_stream import register_firered_vad_stream
from transformers import AutoModel

register_firered_vad_stream()
model = AutoModel.from_pretrained(
    "MigoXV/firered-vad-stream",
    use_safetensors=True,
)

Downloads last month: 26

Safetensors

Model size

568k params

Tensor type

F32

Inference Providers NEW

Voice Activity Detection

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for MigoXV/firered-vad-stream

Base model

FireRedTeam/FireRedVAD

Finetuned

(3)

this model