How to do fine-tuning for Nvidia - NEMO canary-1b models

[kurianbenoy]

Is your feature request related to a problem? Please describe.

• Can you explain, how to fine-tune Nvidia latest ASR models like Canary 1B and NeMo Parakeet ASR Models.

https://nvidia.github.io/NeMo/blogs/2024/2024-02-canary/ https://nvidia.github.io/NeMo/blogs/2024/2024-01-parakeet/

Describe the solution you'd like

• A notebook/code block explaining how to fine-tune Nvidia Canary models for low-resource Indic languages like Malayalam.

[titu1994]

For Parakeet, we have several tutorials in the Nemo tutorials directory - here is one to get started - https://github.com/NVIDIA/NeMo/blob/main/tutorials/asr/ASR_CTC_Language_Finetuning.ipynb

Most of the concepts will apply to Parakeet CTC / RNNT/ TDT.

For Canary, it is a very new model, we are still adding documentation and tutorials for finetuning and scratch training

[titu1994]

Also, if you want a finetune script, you can use this - https://github.com/NVIDIA/NeMo/blob/main/examples/asr/speech_to_text_finetune.py

[kurianbenoy]

Thank @titu1994 for sharing these details