Update app.py

app.py

@@ -1,18 +1,49 @@
 import gradio as gr
-from transformers import pipeline
+import librosa
+import numpy as np
+import torch
 
-pipeline = pipeline(task="image-classification", model="julien-c/hotdog-not-hotdog")
+from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
 
-def predict(input_img):
-    predictions = pipeline(input_img)
-    return input_img, {p["label"]: p["score"] for p in predictions} 
+checkpoint = "mikhail-panzo/zlm-fil-ceb_b64_le5_s8000"
+processor = SpeechT5Processor.from_pretrained(checkpoint)
+model = SpeechT5ForTextToSpeech.from_pretrained(checkpoint)
+vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")
 
-gradio_app = gr.Interface(
-    predict,
-    inputs=gr.Image(label="Select hot dog candidate", sources=['upload', 'webcam'], type="pil"),
-    outputs=[gr.Image(label="Processed Image"), gr.Label(label="Result", num_top_classes=2)],
-    title="Hot Dog? Or Not?",
-)
+def predict(text, speaker):
+    if len(text.strip()) == 0:
+        return (16000, np.zeros(0).astype(np.int16))
 
-if __name__ == "__main__":
-    gradio_app.launch(share=True)
+    inputs = processor(text=text, return_tensors="pt")
+
+    # limit input length
+    input_ids = inputs["input_ids"]
+    input_ids = input_ids[..., :model.config.max_text_positions]
+
+    speaker_embedding = np.load(spe)
+
+    speaker_embedding = torch.tensor(speaker_embedding).unsqueeze(0)
+
+    speech = model.generate_speech(input_ids, speaker_embedding, vocoder=vocoder)
+
+    speech = (speech.numpy() * 32767).astype(np.int16)
+    return (16000, speech)
+
+
+title = "CEB SEQ TTS"
+
+description = """
+This demo system is intended for survey purposes only.
+"""
+
+gr.Interface(
+    fn=predict,
+    inputs=[
+        gr.Text(label="Input Text")
+    ],
+    outputs=[
+        gr.Audio(label="Generated Speech", type="numpy"),
+    ],
+    title=title,
+    description=description,
+).launch()