Model size of pre-trained models

[Tirthankar-iiitb]

How to find model parameters of a pre-trained ESPnet ASR model from model zoo? For example, the SPGIspeech trained model - Shinji_Watanabe/spgispeech_asr_train_asr_conformer6_n_fft512_hop_length256_raw_en_unnorm_bpe5000_valid.acc.ave using ModelDownloader.download_and_unpack(tag)?

[emmastoneviral23]

You can find the parameters used to train an ASR model in the corresponding README file in the model zoo. The most important parameter is the embedding dimension (D). The embedding dimension is the number of unique frames that the model is trained on. A larger embedding dimension will generally improve the model's performance, but will require more time to train. You can check the performance of many models in the model zoo on the model comparison page.

[emmastoneviral23]

The model might be in C, C++, or Python. There are also open-source implementations of models in Torch (don't know if they come with pretrained parameters).

[pakkaujala2]

How to implement the code in C and C++? can you please guide me?