Upload 2 files

furry/kaji/model/kaji/config.py

@@ -0,0 +1,75 @@
+from pytorch_lightning.callbacks import LearningRateMonitor, ModelCheckpoint
+
+from fish_diffusion.datasets.audio_folder import AudioFolderDataset
+
+_base_ = [
+    "./_base_/archs/diff_svc_v2.py",
+    "./_base_/trainers/base.py",
+    "./_base_/schedulers/warmup_cosine_finetune.py",
+    "./_base_/datasets/audio_folder.py",
+]
+
+speaker_mapping = {
+    "kaji_astro_merc": 0,
+    "kaji_astro_delta": 1,
+    "kaji_astro_ksi": 2,
+    "m4_alto_01": 3,
+    "m4_bass_01": 4,
+    "m4_soprano_01": 5,
+    "m4_tenor_01": 6,
+    "m4_tenor_07": 7,
+}
+
+train_datasets = AudioFolderDataset.get_datasets_from_subfolder("dataset/train", speaker_mapping)  # Build datasets manually.
+valid_datasets = AudioFolderDataset.get_datasets_from_subfolder("dataset/valid", speaker_mapping)  # Build datasets manually.
+
+dataset = dict(
+    train=dict(
+        _delete_=True,  # Delete the default train dataset
+        type="ConcatDataset",
+        datasets=train_datasets,
+        # Are there any other ways to do this?
+        collate_fn=AudioFolderDataset.collate_fn,
+    ),
+    valid=dict(
+        _delete_=True,  # Delete the default valid dataset
+        type="ConcatDataset",
+        datasets=valid_datasets,
+        collate_fn=AudioFolderDataset.collate_fn,
+    ),
+)
+
+model = dict(
+    speaker_encoder=dict(
+        input_size=len(speaker_mapping),
+    ),
+    text_encoder=dict(
+        type="NaiveProjectionEncoder",
+        input_size=256,
+        output_size=256,
+    ),
+)
+
+preprocessing = dict(
+    text_features_extractor=dict(
+        type="ChineseHubertSoft",
+        pretrained=True,
+        gate_size=15,
+    ),
+    pitch_extractor=dict(
+        type="ParselMouthPitchExtractor",
+    ),
+)
+
+# The following trainer val and save checkpoints every 1000 steps
+trainer = dict(
+    val_check_interval=5000,
+    callbacks=[
+        ModelCheckpoint(
+            filename="{epoch}-{step}-{valid_loss:.2f}",
+            every_n_train_steps=5000,
+            save_top_k=-1,
+        ),
+        LearningRateMonitor(logging_interval="step"),
+    ],
+)

furry/kaji/model/kaji/model.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd55f5cbd0163e85ccee93f4a7705d36f2eb5459c48b295f24facfe7f22820bc
+size 717795849