feat: add finetuned whisper-large-v3

app.py

@@ -2,6 +2,7 @@ import os
 
 import gradio as gr
 import torch
+from faster_whisper import WhisperModel
 from huggingface_hub import snapshot_download
 from nemo.collections.asr.models import ASRModel
 from omegaconf import OmegaConf
@@ -20,6 +21,13 @@ def load_model(model_id: str):
     return asr_model
 
 
+def load_faster_whisper(model_id: str):
+    # model_dir = snapshot_download(model_id)
+    # model_ckpt_path = os.path.join(model_dir, "model.nemo")
+    model = WhisperModel(model_id)
+    return model
+
+
 OmegaConf.register_new_resolver("load_model", load_model)
 
 models_config = OmegaConf.to_object(OmegaConf.load("configs/models.yaml"))

configs/models.yaml

@@ -1,2 +1,4 @@
+whisper-large-v3-finetune:
+  model: kenkone/kenkone-whisper-large-v3-ct2
 parakeet-0.6b-custom-tokenizer:
   model: ${load_model:kenkone/parakeet-ctc-0.6b-custom-tokenizer}

requirements.txt

@@ -1,4 +1,5 @@
 omegaconf
 git+https://github.com/NVIDIA/NeMo.git@main#egg=nemo_toolkit[asr]
 IPython
-numpy==1.*
+numpy==1.*
+faster-whisper