Fix data logger

app.py

@@ -4,7 +4,23 @@ from datetime import datetime
 from enum import Enum
 from ukrainian_tts.tts import TTS, Stress, Voices
 from torch.cuda import is_available
-from os import environ
+from os import getenv
+from data_logger import log_data
+from threading import Thread
+from queue import Queue
+
+
+def check_thread(logging_queue: Queue):
+    logging_callback = log_data(hf_token=getenv("HF_API_TOKEN"), dataset_name="uk-tts-output", private=True)
+    while True:
+        item = logging_queue.get()
+        logging_callback(item)
+
+if getenv("HF_API_TOKEN") is not None:
+    log_queue = Queue()
+    t = Thread(target=check_thread, args=(log_queue,))
+    t.start()
+
 
 class StressOption(Enum):
     AutomaticStress = "Автоматичні наголоси (за словником) 📖"
@@ -53,18 +69,13 @@ def tts(text: str, voice: str, stress: str):
         text if len(text) < text_limit else text[0:text_limit]
     )  # mitigate crashes on hf space
     
+    if getenv("HF_API_TOKEN") is not None:
+        log_queue.put([text, speaker_name, stress_selected, str(datetime.utcnow())])
 
     with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as fp:
         _, text = ukr_tts.tts(text, speaker_name, stress_selected, fp)
         return fp.name, text
 
-if environ["HF_API_TOKEN"] is None:
-    print("Using default flagging.")
-    flagging_callback = gr.CSVLogger()
-else:
-    print("Using HuggingFace dataset saver.")
-    flagging_callback = gr.HuggingFaceDatasetSaver(hf_token=environ["HF_API_TOKEN"], dataset_name="uk-tts-output", private=True)
-
 
 with open("README.md") as file:
     article = file.read()
@@ -122,9 +133,6 @@ iface = gr.Interface(
             VoiceOption.Lada.value,
             StressOption.AutomaticStress.value,
         ],
-    ],
-    allow_flagging="auto",
-    flagging_callback=flagging_callback,
-    flagging_options=None
+    ]
 )
 iface.launch(enable_queue=True)

data_logger.py

@@ -0,0 +1,41 @@
+from gradio import utils
+import os
+import csv
+import huggingface_hub
+
+def log_data(hf_token: str, dataset_name: str, private=True):
+    path_to_dataset_repo = huggingface_hub.create_repo(
+            name=dataset_name,
+            token=hf_token,
+            private=private,
+            repo_type="dataset",
+            exist_ok=True,
+        )
+    flagging_dir = "flagged"
+    dataset_dir = os.path.join(flagging_dir, dataset_name)
+    repo = huggingface_hub.Repository(
+            local_dir=dataset_dir,
+            clone_from=path_to_dataset_repo,
+            use_auth_token=hf_token,
+        )
+    repo.git_pull(lfs=True)
+    log_file = os.path.join(dataset_dir, "data.csv")
+
+    def log_function(data):
+        repo.git_pull(lfs=True)
+
+        with open(log_file, "a", newline="", encoding="utf-8") as csvfile:
+            writer = csv.writer(csvfile)
+
+            csv_data = data
+            writer.writerow(utils.sanitize_list_for_csv(csv_data))
+
+        with open(log_file, "r", encoding="utf-8") as csvfile:
+            line_count = len([None for row in csv.reader(csvfile)]) - 1
+
+        repo.push_to_hub(commit_message="Flagged sample #{}".format(line_count))
+
+        return line_count
+
+    return log_function
+