Repository navigation
Need An Easy Inference Sample #11
Description
Activity
If you want to perform customizable inference, such as in style transfer, you have two options:
- You can replace the
item_nameused ininpwith the information commented out in the file. For example in style_transfer.py:
# 'text_gen': , # 'note_gen': , # 'note_dur_gen': , # 'note_type_gen':, # 'text_in': , # 'note_in': , # 'note_dur_in': , # 'note_type_in':, # 'ref_audio': , # 'ph_durs':
- Alternatively, you can refer to the GTSinger metadata format to create metadata for your data. This ensures that each WAV file possesses the required attributes. Then, use item_name to facilitate the inference process.
If you want to perform large-scale inference using OpenCPOP data, please follow GTSinger's guidelines to binarize the data. After that, run the following command:
CUDA_VISIBLE_DEVICES=$GPU python tasks/run.py --config egs/sdlm.yaml --exp_name SDLM --inferReacted by 葉軒瑜(Yeh, Hsuan-Yu)- You can replace the
I second this as well. If TCSinger could make a Google Colab tutorial or simple inference sample that would be greatly appreciated!
Reacted by 葉軒瑜(Yeh, Hsuan-Yu)If you want to perform customizable inference, such as in style transfer, you have two options:
- You can replace the
item_nameused ininpwith the information commented out in the file. For example in style_transfer.py:
'text_gen': ,
'note_gen': ,
'note_dur_gen': ,
'note_type_gen':,
'text_in': ,
'note_in': ,
'note_dur_in': ,
'note_type_in':,
'ref_audio': ,
'ph_durs':
- Alternatively, you can refer to the GTSinger metadata format to create metadata for your data. This ensures that each WAV file possesses the required attributes. Then, use item_name to facilitate the inference process.
If you want to perform large-scale inference using OpenCPOP data, please follow GTSinger's guidelines to binarize the data. After that, run the following command:
CUDA_VISIBLE_DEVICES=$GPU python tasks/run.py --config egs/sdlm.yaml --exp_name SDLM --infer
How the 2nd way knows what to infer? like option 1 you give the audio / datas so it gets input and know what to output, but im confused with the 2nd way
- You can replace the
If you want to perform customizable inference, such as in style transfer, you have two options:
- You can replace the
item_nameused ininpwith the information commented out in the file. For example in style_transfer.py:
'text_gen': ,
'note_gen': ,
'note_dur_gen': ,
'note_type_gen':,
'text_in': ,
'note_in': ,
'note_dur_in': ,
'note_type_in':,
'ref_audio': ,
'ph_durs':
- Alternatively, you can refer to the GTSinger metadata format to create metadata for your data. This ensures that each WAV file possesses the required attributes. Then, use item_name to facilitate the inference process.
If you want to perform large-scale inference using OpenCPOP data, please follow GTSinger's guidelines to binarize the data. After that, run the following command:
CUDA_VISIBLE_DEVICES=$GPU python tasks/run.py --config egs/sdlm.yaml --exp_name SDLM --infer
How the 2nd way knows what to infer? like option 1 you give the audio / datas so it gets input and know what to output, but im confused with the 2nd way
For this way, model will find the item_name in the metadata and get all information needed.
- You can replace the
Hello,
I would like to run inference using your provided pretrained model without additional training or data binarization. However, when I tried to run it, I encountered many required files in the
data/directory. On the other hand, there are no clear examples explaining input information format. For instance, instyle_transfer.py, in theexample_runfunction, it does not specify the expected format of information ininp.Could someone share your experience or provide guidance on this?
Additionally, if I want to use my own data (e.g., annotated data from OpenCpop) for inference with the pretrained model, what steps should I follow?
Thank you!