BriefGPT.xyz
Sep, 2023
仅用于困难音频的大模型：高效推理的样本相关私语模型选择
Big model only for hard audios: Sample dependent Whisper model selection for efficient inferences
HTML
PDF
Hugo Malard, Salah Zaiem, Robin Algayres
TL;DR
基于自动语音识别（ASR）领域中模型尺寸和推理成本的观察，通过训练一个决策模块，可以在测试数据的大部分模块上使用最小的足够模型达到良好的转录效果，从而实现了相当大的计算节省和性能提升。
Abstract
Recent progress in
automatic speech recognition
(
asr
) has been coupled with a substantial increase in the
model sizes
, which may now conta
→