When recognizing Chinese, small model is correct, large model is mistake.

[wmlgl]

In the large model , “春到神舟，山河锦绣，龙藤四海，万象蓬勃。各位观众，晚上好，晚上好，今天是2月10号星期六，农历争月初一“ recognized to "中文字幕由 Amara.org 社群提供".

A correct result:

$ whisper  ./test130.wav
/home/yangyp/.conda/envs/whisper/lib/python3.9/site-packages/torch/cuda/__init__.py:141: UserWarning: CUDA initialization: The NVIDIA driver on your system is too old (found version 10020). Please update your GPU driver by downloading and installing a new version from the URL: http://www.nvidia.com/Download/index.aspx Alternatively, go to: https://pytorch.org to install a PyTorch version that has been compiled with your version of the CUDA driver. (Triggered internally at ../c10/cuda/CUDAFunctions.cpp:108.)
  return torch._C._cuda_getDeviceCount() > 0
/home/yangyp/.conda/envs/whisper/lib/python3.9/site-packages/whisper/transcribe.py:126: UserWarning: FP16 is not supported on CPU; using FP32 instead
  warnings.warn("FP16 is not supported on CPU; using FP32 instead")
Detecting language using up to the first 30 seconds. Use `--language` to specify the language
Detected language: Chinese
[00:00.000 --> 00:21.600] 《春到神舟,山河锦绣》
[00:21.600 --> 00:24.240] 龙藤四海,万象彭博
[00:24.240 --> 00:26.240] 各位观众,晚上好
[00:26.240 --> 00:30.080] 今天是2月10号星期六,农历争月初一
[00:30.080 --> 00:31.680] 新闻联播给您
[00:31.680 --> 00:33.280] 拜年了
[00:33.280 --> 00:37.680] 祝福您龙藤虎月,何家幸福,世世顺
[00:37.680 --> 00:42.720] 祝愿我们伟大的祖国,国泰民安,繁荣昌盛,夜夜星
[00:42.720 --> 00:44.720] 欢迎收看新闻联播节目
[00:44.720 --> 00:47.680] 首先为您介绍今天节目的主要内容
[00:47.680 --> 00:50.160] 习近平总书记说
[00:50.160 --> 00:53.840] 大家对新的一年要充满信心,把日子过得更好
[00:54.160 --> 00:56.400] 春节里的中国喜庆祥可
[00:56.400 --> 00:59.200] 各地的人们在丰富多彩的新春活动中
[00:59.200 --> 01:02.400] 感受幸福生活凝聚奋进力量
[01:02.400 --> 01:07.520] 春运以来,全社会跨区域人员流动量近33亿人次
[01:07.520 --> 01:11.360] 总台龙年春晚,亮点分成创意十足
[01:11.360 --> 01:15.840] 新媒体直播规模和海外传播量均创历史新高
[01:15.840 --> 01:20.960] 多国领导人和国际组织负责人向中国人民恭贺新春
[01:20.960 --> 01:24.480] 多个国家举办精彩活动,欢庆中国春节
[01:24.480 --> 01:26.480] 以下来看,详细报道
[01:28.880 --> 01:30.080] 春节是

In the large model, the first line is mistake.

$ whisper --model large ./test130.wav
/home/yangyp/.conda/envs/whisper/lib/python3.9/site-packages/torch/cuda/__init__.py:141: UserWarning: CUDA initialization: The NVIDIA driver on your system is too old (found version 10020). Please update your GPU driver by downloading and installing a new version from the URL: http://www.nvidia.com/Download/index.aspx Alternatively, go to: https://pytorch.org to install a PyTorch version that has been compiled with your version of the CUDA driver. (Triggered internally at ../c10/cuda/CUDAFunctions.cpp:108.)
  return torch._C._cuda_getDeviceCount() > 0
/home/yangyp/.conda/envs/whisper/lib/python3.9/site-packages/whisper/transcribe.py:126: UserWarning: FP16 is not supported on CPU; using FP32 instead
  warnings.warn("FP16 is not supported on CPU; using FP32 instead")
Detecting language using up to the first 30 seconds. Use `--language` to specify the language
Detected language: Chinese
[00:00.000 --> 00:29.980] 中文字幕由 Amara.org 社群提供
[00:30.140 --> 00:33.020] 新闻联播给您拜年了
[00:33.020 --> 00:37.400] 祝福您龙腾虎跃 阖家幸福 世事顺
[00:37.400 --> 00:42.420] 祝愿我们伟大的祖国国泰民安 繁荣昌盛 夜夜兴
[00:42.420 --> 00:44.660] 欢迎收看新闻联播节目
[00:44.660 --> 00:47.400] 首先为您介绍今天节目的主要内容
[00:47.400 --> 00:50.060] 习近平总书记说
[00:50.060 --> 00:53.680] 大家对新的一年要充满信心 把日子过得更好
[00:53.680 --> 00:56.080] 春节里的中国喜庆祥和
[00:56.080 --> 00:58.940] 各地的人们在丰富多彩的新春活动中
[00:58.940 --> 01:02.060] 感受幸福生活凝聚奋进力量
[01:02.060 --> 01:06.760] 春运以来全社会跨区域人员流动量近33亿人次
[01:06.760 --> 01:11.040] 总台龙年春晚亮点分成创意十足
[01:11.040 --> 01:15.380] 新媒体直播规模和海外传播量均创历史新高
[01:15.380 --> 01:20.440] 多国领导人和国际组织负责人向中国人民恭贺新春
[01:20.440 --> 01:24.200] 多个国家举办精彩活动欢庆中国春节
[01:24.200 --> 01:26.080] 以下来看详细报道
[01:28.940 --> 01:30.020] 春节是

test130.zip

[ryanheise]

This is called a hallucination. Would you mind sharing it here: #1873 ? If you search the discussions for "hallucination", you will also find many solutions/workarounds to the problem.