Update app.py

app.py

@@ -1,52 +1,67 @@
-import gradio as gr
+import logging
+from typing import cast
 from pdfminer.high_level import extract_text
-from balacoon_tts import TTS
 from huggingface_hub import hf_hub_download, list_repo_files
+import gradio as gr
+from balacoon_tts import TTS
+
+# Global TTS module, initialized from a selected model
+tts = None
 
 def read_pdf(file):
-    with open(file.name, "rb") as f:
-        text = extract_text(f)
+    text = extract_text(file.name)
     return text
 
-def set_model(model_name_str):
-    model_path = hf_hub_download(repo_id="balacoon/tts", filename=model_name_str)
-    global tts
-    tts = TTS(model_path)
-    speakers = tts.get_speakers()
-    speaker_dropdown.choices = speakers
-    speaker_dropdown.update(default=speakers[0] if speakers else None)
-
 def synthesize_audio(file, model_name_str, speaker_str):
     text_str = read_pdf(file)
     if len(text_str) > 1024:
         text_str = text_str[:1024]
     samples = tts.synthesize(text_str, speaker_str)
-    return (tts.get_sampling_rate(), samples)
+    return (samples, tts.get_sampling_rate())
 
 def main():
-    repo_files = list_repo_files(repo_id="balacoon/tts")
-    model_files = [x for x in repo_files if x.endswith("_cpu.addon")]
-    model_name_dropdown = gr.inputs.Dropdown(label="Model", choices=model_files)
-    speaker_dropdown = gr.inputs.Dropdown(label="Speaker", choices=[])
+    logging.basicConfig(level=logging.INFO)
+    with gr.Interface(fn=None, title="PDF TO SPEECH CONVERTER", layout="rows", debug=True) as iface:
+        gr.Markdown(
+            """
+            <h1 align="center">PDF TO SPEECH CONVERTER</h1>
+            1. Insert a PDF 
+            2. Select the model to synthesize with
+            3. Select speaker
+            4. Hit "Generate" and listen to the result!
+            When you select a model for the first time, it may take some time to download it.
+            This project is designed to bring the joy of reading without the hassle of looking over.
+            If you want an audiobook, you've got it!
+            """
+        )
+
+        file_input = gr.inputs.File(label="Upload PDF")
+
+        model_files = list_repo_files(repo_id="balacoon/tts")
+        model_name_dropdown = gr.inputs.Dropdown(label="Model", choices=model_files)
 
-    file_input = gr.inputs.File(label="Select a PDF File", type="file")
-    text = gr.outputs.Textbox()
+        def set_model(model_name_str: str):
+            model_path = hf_hub_download(repo_id="balacoon/tts", filename=model_name_str)
+            global tts
+            tts = TTS(model_path)
+            speakers = tts.get_speakers()
+            default_speaker = speakers[-1]
+            return speakers, default_speaker
 
-    audio = gr.outputs.Audio(label="Generated Audio")
+        model_name_dropdown.set_action(set_model)
 
-    iface = gr.Interface(
-        fn=synthesize_audio,
-        inputs=[file_input, model_name_dropdown, speaker_dropdown],
-        outputs=audio,
-        title="PDF TO SPEECH CONVERTER",
-        layout="rows",
-        debug=True
-    )
+        speaker_dropdown = gr.inputs.Dropdown(label="Speaker", choices=[])
 
-    model_name_dropdown.set_action(set_model)
+        audio = gr.outputs.Audio(label="Generated Audio", type="numpy")
 
-    iface.launch()
+        def generate_audio(file, model_name_str, speaker_str):
+            return synthesize_audio(file, model_name_str, speaker_str)
 
+        iface.add_input(file_input)
+        iface.add_input(model_name_dropdown)
+        iface.add_input(speaker_dropdown)
+        iface.add_output(audio)
+        iface.run()
 
 if __name__ == "__main__":
     main()