Update app.py

app.py

@@ -10,6 +10,11 @@ import zipfile
 from io import BytesIO
 import re
 
+from pydub import AudioSegment
+
+final_audio = AudioSegment.empty()
+
+
 # Voice model
 VOICE_MODEL = "tts_models/en/vctk/vits"
 
@@ -106,87 +111,88 @@ SPEAKER_METADATA = {
 	    273: { "age": 18, "gender": "F", "accent": "English"}
 	}
 
+def clean_text(text):
+    # Remove hyperlinks
+    return re.sub(r'http[s]?://\S+', '', text)
+
+def extract_paragraphs_from_docx(docx_file):
+    document = Document(docx_file.name)
+    paragraphs = [p.text.strip() for p in document.paragraphs if p.text.strip()]
+    return [clean_text(p) for p in paragraphs]
+
+def list_speaker_choices():
+    return [f"{sid} | {meta['gender']} | {meta['accent']}" for sid, meta in speaker_metadata.items()]
 
+def get_speaker_id_from_label(label):
+    return label.split('|')[0].strip()
 
-# Static list of speakers for dropdown
-SPEAKER_CHOICES = [
-    f"{sid} - {data['gender']} - {data['accent']} (Age {data['age']})"
-    for sid, data in SPEAKER_METADATA.items()
-]
-
-# VCTK model (multi-speaker)
-MODEL_NAME = "tts_models/en/vctk/vits"
-tts = TTS(model_name=MODEL_NAME, progress_bar=False, gpu=False)
-
-# Extract plain text from docx, ignoring hyperlinks
-def extract_text_ignoring_hyperlinks(docx_file):
-    doc = Document(docx_file.name)
-    text_blocks = []
-    for para in doc.paragraphs:
-        # Remove hyperlinks using regex or by inspecting runs
-        if para.text.strip():
-            clean_text = re.sub(r'https?://\S+', '', para.text)
-            text_blocks.append(clean_text.strip())
-    return text_blocks
-
-# Generate sample audio for preview
-def generate_sample_audio(sample_text, selected_speaker):
-    if not sample_text.strip():
-        raise gr.Error("Sample text cannot be empty.")
-    sid = selected_speaker.split(" ")[0]  # Extract speaker ID
+def generate_sample_audio(sample_text, speaker_label):
+    speaker_id = get_speaker_id_from_label(speaker_label)
+    model = TTS("tts_models/multilingual/multi-dataset/your_model")
     with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp_wav:
-        tts.tts_to_file(text=sample_text, speaker="p"+sid, file_path=tmp_wav.name)
+        model.tts_to_file(text=sample_text, speaker="p"+speaker_id, file_path=tmp_wav.name)
         return tmp_wav.name
 
-# Main conversion function
-def docx_to_zipped_wavs(doc_file, selected_speaker):
-    sid = selected_speaker.split(" ")[0]
-    paragraphs = extract_text_ignoring_hyperlinks(doc_file)
-    audio_files = []
+def generate_audio(docx_file, speaker_label):
+    speaker_id = get_speaker_id_from_label(speaker_label)
+    model = TTS("tts_models/multilingual/multi-dataset/your_model")
+
+    paragraphs = extract_paragraphs_from_docx(docx_file)
+    combined_audio = AudioSegment.empty()
+    temp_files = []
+
     try:
-        for i, para in enumerate(paragraphs):
-            if not para:
-                continue
-            with tempfile.NamedTemporaryFile(suffix=f"_{i}.wav", delete=False) as tmp_wav:
-                tts.tts_to_file(text=para, speaker="p"+sid, file_path=tmp_wav.name)
-                audio_files.append(tmp_wav.name)
+        for idx, para in enumerate(paragraphs):
+            tmp = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
+            model.tts_to_file(text=para, speaker="p"+speaker_id, file_path=tmp.name)
+            audio_chunk = AudioSegment.from_wav(tmp.name)
+            combined_audio += audio_chunk
+            temp_files.append(tmp.name)
+            tmp.close()
+
     except Exception as e:
-        print("Connection interrupted, returning partial result.", str(e))
-
-    # Zip the results
-    zip_buffer = BytesIO()
-    with zipfile.ZipFile(zip_buffer, "w") as zipf:
-        for wav_path in audio_files:
-            zipf.write(wav_path, arcname=os.path.basename(wav_path))
-    zip_buffer.seek(0)
-
-    # Save the zip temporarily for download
-    final_zip = tempfile.NamedTemporaryFile(delete=False, suffix=".zip")
-    final_zip.write(zip_buffer.read())
-    final_zip.close()
-    return final_zip.name
-
-# Gradio UI
-with gr.Blocks() as interface:
-    gr.Markdown("""# Multi-Paragraph Voiceover Generator
-Upload a `.docx` file and convert each paragraph to audio. You can also try a short sample first.
-""")
+        print("Generation interrupted. Saving partial output.", e)
 
-    with gr.Row():
-        #sample_text = gr.Textbox(label="Sample Text (Max 500 chars)", max_lines=4, lines=3, max_length=500)
-        speaker_dropdown = gr.Dropdown(label="Select Speaker", choices=SPEAKER_CHOICES, value=SPEAKER_CHOICES[0])
+    output_dir = tempfile.mkdtemp()
+    final_output_path = os.path.join(output_dir, "final_output.wav")
+    combined_audio.export(final_output_path, format="wav")
+
+    zip_path = os.path.join(output_dir, "output.zip")
+    with zipfile.ZipFile(zip_path, 'w') as zipf:
+        zipf.write(final_output_path, arcname="final_output.wav")
 
-    # sample_button = gr.Button("Generate Sample Audio")
-    # sample_audio = gr.Audio(label="Sample Audio", type="filepath")
+    for f in temp_files:
+        os.remove(f)
+
+    return zip_path
+
+# --- UI ---
+speaker_choices = list_speaker_choices()
+
+with gr.Blocks() as demo:
+    gr.Markdown("## 📄 TTS Voice Generator with Paragraph-Wise Processing")
+
+    with gr.Row():
+        speaker_dropdown = gr.Dropdown(label="Select Voice", choices=speaker_choices)
 
     with gr.Row():
-        docx_input = gr.File(label="Upload .docx File", type="filepath")
-        convert_button = gr.Button("Generate WAV Zip")
+        sample_textbox = gr.Textbox(label="Enter Sample Text (Max 500 characters)", max_lines=5, max_chars=500)
+        sample_button = gr.Button("Generate Sample")
+        clear_button = gr.Button("Clear Sample")
 
-    final_output = gr.File(label="Download ZIP of WAVs")
+    sample_audio = gr.Audio(label="Sample Output", type="filepath")
 
-    # sample_button.click(fn=generate_sample_audio, inputs=[sample_text, speaker_dropdown], outputs=sample_audio)
-    convert_button.click(fn=docx_to_zipped_wavs, inputs=[docx_input, speaker_dropdown], outputs=final_output)
+    sample_button.click(fn=generate_sample_audio, inputs=[sample_textbox, speaker_dropdown], outputs=[sample_audio])
+    clear_button.click(fn=lambda: None, inputs=[], outputs=[sample_audio])
+
+    with gr.Row():
+        docx_input = gr.File(label="Upload DOCX File", file_types=[".docx"])
+        generate_button = gr.Button("Generate Full Audio")
+        download_output = gr.File(label="Download Output Zip")
+
+    generate_button.click(fn=generate_audio, inputs=[docx_input, speaker_dropdown], outputs=[download_output])
 
 if __name__ == "__main__":
-    interface.launch()
+    demo.launch()
+
+