update script for ov

scripts/run_whisper_finetuned_with_punc_ov.py

@@ -43,7 +43,7 @@ def transcribe_file(
 def load_model(device):
     device = "GPU"  # GPU can be used as well
     # model_path = r"D:\yujuan\yoyo-translator-win\models\whisper-large-v3-turbo-int8"
-    model_path = r"D:\yujuan\models\whisper-turbo-25000-int8p\whisper-turbo-25000-int8p"
+    model_path = r"D:\yujuan\models\whisper-turbo-39000-int8p\whisper-turbo-39000-int8p"
     punc_model = r"D:\yujuan\models\funasr_ct\ct-punc"
 
     t0 = time.time()
@@ -94,6 +94,75 @@ def run_recordings():
         rows.append([audio.name, round(t, 3), text, d, round(nd,3), diff])
     save_csv("csv/finetune_whisper_with_punc.csv", rows)
 
+def run_test_dataset():
+    from scripts.asr_utils import read_dataset
+    device = "GPU"
+    lang = "<|zh|>"
+    asr, punc = load_model(device)
+
+    test_data = Path("../tests/test_data/dataset.txt")
+    audio_parent = Path("../tests/test_data/")
+    rows = [["file_name", "time", "inference_result"]]
+    result_list = []
+    count = 0
+    try:
+        for audio_path, sentence, duration in read_dataset(test_data):
+            count += 1
+            print(f"processing {count}: {audio_path}")
+
+            text, t = inference(audio_parent/audio_path, asr, punc, lang)
+            print("inference time:", t)
+            print(text)
+            result_list.append({
+                "index": count,
+                "audio_path": audio_path,
+                "reference": sentence,
+                "duration": duration,
+                "inference_time": round(t, 3),
+                "inference_result": text
+            })
+    except Exception as e:
+        print(e)
+    except KeyboardInterrupt as e:
+        print(e)
+    import json
+    with open("csv/whisper_finetune_ov_results.json", "w", encoding="utf-8") as f:
+        json.dump(result_list, f, ensure_ascii=False, indent=2)
+
+
+def run_test_emilia():
+    from scripts.asr_utils import read_emilia
+    device = "GPU"
+    lang = "<|zh|>"
+    asr, punc = load_model(device)
+    parent = Path("../tests/test_data/ZH-B000000")
+    result_list = []
+    count = 0
+    try:
+        for audio_path, sentence, duration in read_emilia(parent, count_limit=5000):
+            count += 1
+            print(f"processing {count}: {audio_path.name}")
+
+            text, t = inference(audio_path, asr, punc, lang)
+            print("inference time:", t)
+            print(text)
+            result_list.append({
+                "index": count,
+                "audio_path": audio_path.name,
+                "reference": sentence,
+                "duration": duration,
+                "inference_time": round(t, 3),
+                "inference_result": text
+            })
+    except Exception as e:
+        print(e)
+    except KeyboardInterrupt as e:
+        print(e)
+    import json
+    with open("csv/whisper_finetune_emilia_ov_results.json", "w", encoding="utf-8") as f:
+        json.dump(result_list, f, ensure_ascii=False, indent=2)
+
+
 if __name__ == "__main__":
     # main()
-    run_recordings()
+    run_test_emilia()