Upload 2 files

model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ae4c6d08511f62090c5fe5a02cd9effc57d147e8315f7fd477b20a20ce4d13b
+size 113835419

pipeline.py

@@ -0,0 +1,42 @@
+from typing import Tuple
+import subprocess
+
+from torch import no_grad, package
+import numpy as np
+import os
+
+
+
+
+
+class PreTrainedPipeline():
+    def __init__(self, path: str):
+        # Install espeak-ng
+        subprocess.run("apt-get update -y && apt-get install espeak-ng -y", shell=True, 
+            universal_newlines=True, start_new_session=True)
+
+        # Init model
+        model_path = os.path.join(path, "model.pt")
+        importer = package.PackageImporter(model_path)
+        synt = importer.load_pickle("tts_models", "model")
+        self.synt = synt
+
+        self.tts_kwargs = {
+            "speaker_name": "uk",
+            "language_name": "uk",
+        }
+
+        self.sampling_rate = self.synt.output_sample_rate
+
+    def __call__(self, inputs: str) -> Tuple[np.array, int]:
+        """
+        Args:
+            inputs (:obj:`str`):
+                The text to generate audio from
+        Return:
+            A :obj:`np.array` and a :obj:`int`: The raw waveform as a numpy array, and the sampling rate as an int.
+        """
+        with no_grad():
+            waveforms = self.synt.tts(inputs, **self.tts_kwargs)
+            waveforms = np.array(waveforms, dtype=np.float32)
+        return waveforms, self.sampling_rate