Unverified50% confidenceBenchmarkExact time
Distil-Whisper通过蒸馏将Whisper large-v2压缩为仅2层解码器的版本，推理速度提升6倍，在分布外测试集上仅损失1%的词错率
1
Sources
50%
Confidence
Long-term
Relevance
8/26/2026
First Seen
Sources
VoiceGecko：完全本地运行的开源语音转文字工具
Related Entities
Related Claims
UnverifiedIn the Prefill phase, performance differences among the three model types are within the noise range; in the Decode phase, performance is actually about 1.4% faster after applying the patch.71% similarUnverifiedfaster-whisper基于CTranslate2优化推理引擎，在保持相同准确率的同时将推理速度提升4倍、内存占用降低3倍70% similarUnverified本地模型识别准确率取决于所选模型规模，Whisper tiny/base 体积仅几十 MB 精度有限，Whisper medium/large 可达数 GB 精度更高但算力要求更高69% similarUnverifiedModel compression techniques such as INT8 quantization and channel pruning can increase inference speed by 2–4 times with a precision loss of less than 1% mAP69% similarUnverified研究表明在干净测试集上得分相近的两个模型面对带噪声真实输入时性能差距可能扩大数倍63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/802486API
curl https://kongchang.com/api/v1/knowledge/claims/802486MCP
get_claim(id=802486)