Add anime-whisper

gradio_tabs/dataset.py

@@ -177,9 +177,16 @@ def create_dataset_app() -> gr.Blocks:
                     value=False,
                 )
                 hf_repo_id = gr.Dropdown(
-                    ["openai/whisper", "kotoba-tech/kotoba-whisper-v1.1"],
-                    label="HuggingFaceのWhisperモデル",
-                    value="openai/whisper",
+                    [
+                        "openai/whisper-large-v3-turbo",
+                        "openai/whisper-large-v3",
+                        "openai/whisper-large-v2",
+                        "kotoba-tech/kotoba-whisper-v1.1",
+                        "kotoba-tech/kotoba-whisper-v2.1",
+                        "litagin/anime-whisper",
+                    ],
+                    label="HuggingFaceのWhisper repo_id",
+                    value="openai/whisper-large-v3",
                     visible=False,
                 )
                 compute_type = gr.Dropdown(

transcribe.py

@@ -67,8 +67,7 @@ def transcribe_files_with_hf_whisper(
         "num_beams": num_beams,
         "no_repeat_ngram_size": no_repeat_ngram_size,
     }
-    logger.info(f"generate_kwargs: {generate_kwargs}")
-
+    logger.info(f"generate_kwargs: {generate_kwargs}, loading pipeline...")
     pipe = pipeline(
         model=model_id,
         max_new_tokens=128,
@@ -79,6 +78,7 @@ def transcribe_files_with_hf_whisper(
         trust_remote_code=True,
         # generate_kwargs=generate_kwargs,
     )
+    logger.info("Loaded pipeline")
     if initial_prompt is not None:
         prompt_ids: torch.Tensor = pipe.tokenizer.get_prompt_ids(
             initial_prompt, return_tensors="pt"
@@ -195,9 +195,6 @@ if __name__ == "__main__":
             wav_rel_path = wav_file.relative_to(input_dir)
             with open(output_file, "a", encoding="utf-8") as f:
                 f.write(f"{wav_rel_path}|{model_name}|{language_id}|{text}\n")
-    else:
-        if args.hf_repo_id == "":
-            model_id = f"openai/whisper-{args.model}"
     else:
         model_id = args.hf_repo_id
         logger.info(f"Loading HF Whisper model ({model_id})")