phoonnx 0.2.6a2__tar.gz → 0.2.7a1__tar.gz

{phoonnx-0.2.6a2 → phoonnx-0.2.7a1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: phoonnx
-Version: 0.2.6a2
+Version: 0.2.7a1
 Home-page: https://github.com/TigreGotico/phoonnx
 Author: JarbasAi
 Author-email: jarbasai@mailfence.com

{phoonnx-0.2.6a2 → phoonnx-0.2.7a1}/phoonnx/version.py

@@ -1,8 +1,8 @@
 # START_VERSION_BLOCK
 VERSION_MAJOR = 0
 VERSION_MINOR = 2
-VERSION_BUILD = 6
-VERSION_ALPHA = 2
+VERSION_BUILD = 7
+VERSION_ALPHA = 1
 # END_VERSION_BLOCK
 
 VERSION_STR = f"{VERSION_MAJOR}.{VERSION_MINOR}.{VERSION_BUILD}"

{phoonnx-0.2.6a2 → phoonnx-0.2.7a1}/phoonnx.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: phoonnx
-Version: 0.2.6a2
+Version: 0.2.7a1
 Home-page: https://github.com/TigreGotico/phoonnx
 Author: JarbasAi
 Author-email: jarbasai@mailfence.com

{phoonnx-0.2.6a2 → phoonnx-0.2.7a1}/phoonnx_train/preprocess.py

@@ -377,6 +377,16 @@ def phonemize_worker(
     is_flag=True,
     help="Add diacritics to text (phonemizer specific, e.g., to denote stress).",
 )
+@click.option(
+    "--jsonl-audio-path",
+    default=None,
+    help="override audio_path base directory (everything before '/wav') in generated dataset.jsonl"
+)
+@click.option(
+    "--jsonl-audio-spec-path",
+    default=None,
+    help="override audio_norm_path/audio_spec_path base directory (everything before '/cache') in generated dataset.jsonl"
+)
 def cli(
     input_dir: Path,
     output_dir: Path,
@@ -397,6 +407,8 @@ def cli(
     skip_audio: bool,
     debug: bool,
     add_diacritics: bool,
+    jsonl_audio_path: Optional[str],
+    jsonl_audio_spec_path: Optional[str],
 ) -> None:
     """
     Preprocess a TTS dataset (e.g., LJSpeech format) for training a VITS-style model.
@@ -609,6 +621,17 @@ def cli(
                 _LOGGER.warning("Skipping utterance with invalid phoneme_ids before writing: %s", utt.audio_path)
                 continue
 
+            # apply path overrides if needed
+            # this allows pre-processing the dataset in one system and then train in other
+            if jsonl_audio_path:
+                base_path, fname = str(utt.audio_path).split("/wav/")
+                utt.audio_path = Path(f"{jsonl_audio_path}/wav/{fname}")
+            if jsonl_audio_spec_path:
+                base_path, fname = str(utt.audio_norm_path).split("/cache/")
+                utt.audio_norm_path = Path(f"{jsonl_audio_spec_path}/cache/{fname}")
+                base_path, fname = str(utt.audio_spec_path).split("/cache/")
+                utt.audio_spec_path = Path(f"{jsonl_audio_spec_path}/cache/{fname}")
+
             json.dump(
                 utt.asdict(),
                 dataset_file,