Unverified50% confidenceFactExact time
A certain speech recognition model is only 164MB in size and transcribes one hour of audio in about 20 seconds on an ordinary thin-and-light laptop, with a word error rate of 5.21, outperforming mainstream models such as Whisper Large V3 Turbo; it currently supports only English.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/22/2026
First Seen
Valid until: 1/20/2027
Sources
Related Entities
Related Claims
UnverifiedWhisper small模型在现代笔记本电脑的CPU上处理一段10分钟音频通常只需1-3分钟78% similarUnverified在 M1 芯片上运行 Whisper base 模型的转录速度可达音频时长的 10-15 倍，即 1 分钟音频仅需 4-6 秒完成转录77% similarVerifiedThe sampling rate of audio signals is typically 16 kHz or 44.1 kHz, containing tens of thousands of sample points per second, while the output text sequence contains only approximately 3–5 words per second.77% similarUnverifiedWhisper的语音识别模型只能处理16kHz采样率的单声道PCM浮点数据格式的原始音频波形76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1087034API
curl https://kongchang.com/api/v1/knowledge/claims/1087034MCP
get_claim(id=1087034)