How to infer models in the TS section?

[hardikkamboj]

Hello team,

Thanks for proving the models, really appreciate the effort.

I am currently trying to infer on the TS models present in the table under the section "wav2vec2 based models" in the main readme of this repository. However I am unable to load it using the huggingface code (the model here is .pt and not .bin as in the huggingface models). Also, the files downloaded from the repository only contains the josn file and model file, and is missing the config files.

Can you please help on which script I can use to infer these models? (for example english_ts)

[Awaisn25]

Probably too late to comment but for anyone else who is wondering how to perform inference using these models:
Its a simple forward() call on the model with the 2D pytorch tensor of waveform as a parameter. Incase your audio is mono, it will be loaded as 1D tensor so in that case:

import librosa
import torch
data, sr = librosa.load(<path_to_audio>, sr=<sample_rate>) #this is a mono audio
data_p = torch.unsqueeze(torch.from_numpy(data), 0)

model = torch.jit.load(<path_to_model>)
model(data_p)