fixed train and dev speaker overlap

libritts-r-aligned.py

@@ -26,7 +26,7 @@ _PATH = os.environ.get("LIBRITTS_PATH", os.environ.get("HF_DATASETS_CACHE", None
 if _PATH is not None and not os.path.exists(_PATH):
     os.makedirs(_PATH)
 
-_VERSION = "1.0.1"
+_VERSION = "1.1.0"
 
 _CITATION = """\
 @article{koizumi2023libritts,
@@ -133,12 +133,11 @@ class LibriTTSAlign(datasets.GeneratorBasedBuilder):
         data_train = self._create_data([ds_dict["train-clean-100"], ds_dict["train-clean-360"], ds_dict["train-other-500"]])
         data_dev = self._create_data([ds_dict["dev-clean"], ds_dict["dev-other"]])
         data_test = self._create_data([ds_dict["test-clean"], ds_dict["test-other"]])
-        data_all = pd.concat([data_train, data_dev, data_test])
         splits += [
             datasets.SplitGenerator(
                 name="train.all",
                 gen_kwargs={
-                    "ds": data_all,
+                    "ds": data_train,
                 }
             ),
             datasets.SplitGenerator(
@@ -154,13 +153,30 @@ class LibriTTSAlign(datasets.GeneratorBasedBuilder):
                 }
             ),
         ]
-        # move last row for each speaker from data_all to dev dataframe
-        data_dev = data_all.copy()
-        data_dev = data_dev.sort_values(by=["speaker", "audio"])
-        data_dev = data_dev.groupby("speaker").tail(1)
-        data_dev = data_dev.reset_index()
-        # remove last row for each speaker from data_all
-        data_all = data_all[~data_all["audio"].isin(data_dev["audio"])]
+        data_all = pd.concat([data_train, data_dev, data_test])
+        # create a new split which takes one sample from each speaker in data_all and puts it into the dev split
+        # we then remove these samples from data_all
+        speakers = data_all["speaker"].unique()
+        # seed for reproducibility
+        np.random.seed(42)
+        data_dev_all = None
+        for speaker in tqdm(speakers, desc="creating dev split"):
+            data_speaker = data_all[data_all["speaker"] == speaker]
+            if len(data_speaker) < 10:
+                print(f"Speaker {speaker} has only {len(data_speaker)} samples, skipping")
+            else:
+                data_speaker = data_speaker.sample(1)
+                data_all = data_all[data_all["audio"] != data_speaker["audio"].values[0]]
+                if data_dev_all is None:
+                    data_dev_all = data_speaker
+                else:
+                    data_dev_all = pd.concat([data_dev_all, data_speaker])
+        data_all = data_all[data_all["speaker"].isin(data_dev_all["speaker"].unique())]
+        self.speaker2idxs = {}
+        self.speaker2idxs["all"] = {speaker: idx for idx, speaker in enumerate(sorted(list(data_dev_all["speaker"].unique())))}
+        self.speaker2idxs["train"] = {speaker: idx for idx, speaker in enumerate(sorted(list(data_train["speaker"].unique())))}
+        self.speaker2idxs["dev"] = {speaker: idx for idx, speaker in enumerate(sorted(list(data_dev["speaker"].unique())))}
+        self.speaker2idxs["test"] = {speaker: idx for idx, speaker in enumerate(sorted(list(data_test["speaker"].unique())))}
         splits += [
             datasets.SplitGenerator(
                 name="train",
@@ -171,7 +187,7 @@ class LibriTTSAlign(datasets.GeneratorBasedBuilder):
             datasets.SplitGenerator(
                 name="dev",
                 gen_kwargs={
-                    "ds": data_dev,
+                    "ds": data_dev_all,
                 }
             ),
         ]
@@ -234,6 +250,7 @@ class LibriTTSAlign(datasets.GeneratorBasedBuilder):
             entries += add_entries
         if self.empty_textgrids > 0:
             logger.warning(f"Found {self.empty_textgrids} empty textgrids")
+        del self.ds, self.phone_cache, self.phone_converter
         return pd.DataFrame(
             entries,
             columns=[
@@ -247,7 +264,6 @@ class LibriTTSAlign(datasets.GeneratorBasedBuilder):
                 "basename",
             ],
         )
-        del self.ds, self.phone_cache, self.phone_converter
 
     def _create_entry(self, dsi_idx):
         dsi, idx = dsi_idx