初始化项目，由ModelHub XC社区提供模型

Model: kotoba-tech/kotoba-whisper-v1.0
Source: Original Platform

benchmark.py

@@ -0,0 +1,30 @@
+from time import time
+from pprint import pprint
+import torch
+from transformers import pipeline
+from datasets import load_dataset
+
+# config
+generate_kwargs = {"language": "japanese", "task": "transcribe"}
+model_id = "kotoba-tech/kotoba-whisper-v1.0"
+
+# load model
+pipe = pipeline(
+    "automatic-speech-recognition",
+    model=model_id,
+    torch_dtype=torch.float32
+)
+
+test_audio = [
+    "kotoba-whisper-eval/audio/manzai1.wav",
+    "kotoba-whisper-eval/audio/manzai2.wav",
+    "kotoba-whisper-eval/audio/manzai3.wav",
+    "kotoba-whisper-eval/audio/long_interview_1.wav",
+]
+elapsed = {}
+for x in test_audio:
+    start = time()
+    transcription = pipe(x, generate_kwargs=generate_kwargs)
+    elapsed[x] = time() - start
+    pprint(transcription)
+    pprint(elapsed)