Dev.ap/styletts2

.github/workflows/test.yml

@@ -17,7 +17,6 @@ jobs:
       - uses: actions/setup-python@v6
         with:
           python-version: "3.10"
-          cache: "pip"
 
       - uses: awalsh128/cache-apt-pkgs-action@acb598e5ddbc6f68a970c5da0688d2f3a9f04d05 # v1.6.0
         with:
@@ -33,26 +32,37 @@ jobs:
           git fetch origin ${{ github.ref }}
           git checkout FETCH_HEAD
 
-      - name: Install everyvoice
-        run: |
-          cd EveryVoice
-          CUDA_TAG=cpu pip install -r requirements.torch.txt --find-links https://download.pytorch.org/whl/torch_stable.html
-          pip install -e .[dev]
+      - name: Install uv
+        uses: astral-sh/setup-uv@08807647e7069bb48b6ef5acd8ec9567f424441b # v8.1.0
+        with:
+          # Install a specific version of uv.
+          version: "0.11.8"
 
-      - run: pip freeze
+      - name: Install the project
+        run: |
+          cd EveryVoice/everyvoice/model/feature_prediction/FastSpeech2_lightning/
+          uv sync --locked --all-extras --dev
+          uv pip install --editable ../../../..
 
-      - run: pip list
+      - run: |
+          cd EveryVoice/everyvoice/model/feature_prediction/FastSpeech2_lightning/
+          uv pip freeze
+      - run: |
+          cd EveryVoice/everyvoice/model/feature_prediction/FastSpeech2_lightning/
+          uv pip list
 
       - name: Run fs2 unit tests
         run: |
           cd EveryVoice/everyvoice/model/feature_prediction/FastSpeech2_lightning/
-          coverage run -m pytest
-          coverage xml
+          uv run coverage run -m pytest
+          uv run coverage xml
+        env:
+          PYTHONPATH: ${{ github.workspace }}/EveryVoice
 
       - name: Plain text coverage report
         run: |
           cd EveryVoice/everyvoice/model/feature_prediction/FastSpeech2_lightning/
-          coverage report
+          uv run coverage report
 
       - uses: codecov/codecov-action@v5
         with:

fs2/prediction_writing_callback.py

@@ -9,6 +9,9 @@
 import numpy.typing as npt
 import torch
 import torchaudio
+from everyvoice.base_cli.prediction_writing_callback import (
+    BasePredictionWritingCallback,
+)
 from everyvoice.model.feature_prediction.FastSpeech2_lightning.fs2.utils import (
     truncate_basename,
 )
@@ -114,7 +117,7 @@ def get_synthesis_output_callbacks(
     return callbacks
 
 
-class PredictionWritingCallbackBase(Callback):
+class PredictionWritingCallbackBase(BasePredictionWritingCallback):
     def __init__(
         self,
         config: FastSpeech2Config,
@@ -123,31 +126,13 @@ def __init__(
         save_dir: Path,
         include_global_step_in_filename: bool = False,
     ) -> None:
-        super().__init__()
+        super().__init__(
+            save_dir=save_dir,
+            file_extension=file_extension,
+            global_step=global_step,
+            include_global_step_in_filename=include_global_step_in_filename,
+        )
         self.config = config
-        self.file_extension = file_extension
-        self.global_step = f"ckpt={global_step}"
-        self.save_dir = save_dir
-        self.sep = "--"
-        self.include_global_step_in_filename = include_global_step_in_filename
-
-        self.save_dir.mkdir(parents=True, exist_ok=True)
-
-    def get_filename(
-        self,
-        basename: str,
-        speaker: str,
-        language: str,
-    ) -> str:
-        # We don't truncate or alter the filename here because the basename is
-        # already truncated/cleaned in cli/synthesize.py
-        name_parts = [basename, speaker, language, self.file_extension]
-        if self.include_global_step_in_filename:
-            name_parts.insert(-1, self.global_step)
-        path = self.save_dir / self.sep.join(name_parts)
-        # synthesizing spec allows nested outputs so we may need to make subdirs
-        path.parent.mkdir(parents=True, exist_ok=True)
-        return str(path)
 
 
 class ScorerCallback(Callback):
@@ -662,7 +647,7 @@ def __init__(
         self.output_hop_size = (
             sampling_rate_change * vocoder_config.preprocessing.audio.fft_hop_size
         )
-        self.file_extension = self.sep.join(
+        self.file_extension: str = self.sep.join(
             (f"v_ckpt={vocoder_global_step}", self.file_extension)
         )
 

pyproject.toml

@@ -8,7 +8,7 @@ dynamic = ["version"]
 description = "Text-to-Speech Synthesis for the Speech Generation for Indigenous Language Education Small Teams Project"
 license = "MIT"
 readme = "readme.md"
-requires-python = ">=3.10"
+requires-python = ">=3.10, <3.13"
 authors = [
     { name = "Aidan Pine", email = "hello@aidanpine.ca" },
     { name = "Samuel Larkin", email = "Samuel.Larkin@nrc-cnrc.gc.ca" },