Repository navigation
runtime error for audio file in which there is actually no speech (blank voice) #1386
Description
🐛 Bug
To Reproduce
from funasr import AutoModel
model = AutoModel(model="paraformer-zh", model_revision="v2.0.4",
vad_model="fsmn-vad", vad_model_revision="v2.0.4",
punc_model="ct-punc-c", punc_model_revision="v2.0.4",)
result = model.generate(input=common_voice_zh-CN_33658424.mp3.zip,
batch_size_s=300,)
Expected behavior
The above attached audio file is blank voice, so the reasonable output can be result = [{'key': 'xxx', 'text': '', 'timestamp': []}]. However, the above code run into the following RuntimeError:
RuntimeError: Expected tensor for argument #1 'indices' to have one of the following scalar types: Long, Int; but got torch.cuda.DoubleTensor instead (while checking arguments for embedding)
My guess is that asr model output empty text, which however can not be appropriately handled by the current code in the punc model.
Environment
- OS: Linux
- FunASR Version: 1.0.5
- ModelScope Version: 1.12.0
- PyTorch Version: 1.12.0
- How you installed funasr: pip
- Python version: 3.9
- GPU: p40
- CUDA/cuDNN version: 113
Activity
R1ckShi commentedon Feb 27, 2024
thanks for feedback, bug has been fixed, please pull the latest code
OoKyleoO commentedon Mar 1, 2024
Hi, I've pull the latest code and still having the same issue.
Is there any mistake in my code?
self.audio_model.generate(input=[audio], batch_size_s=300, is_final=True, sentence_timestamp=True, hotwords=' '.join(list(self.keywords))jianganghan commentedon Mar 1, 2024
It works for me. You may want to try the latest code with recent changes, eg changes in this pr.
OoKyleoO commentedon Mar 1, 2024
I'll try now. THX!!