mp3 gradio, rss logic

app.py

@@ -10,10 +10,19 @@ import time
 import pymupdf
 import requests
 from pathlib import Path
+from pydub import AudioSegment  # Add this import
+import tempfile
 
 import torch
 from huggingface_hub import InferenceClient
 from kokoro import KModel, KPipeline
+
+# -----------------------------------------------------------------------------
+# to-do 
+# - Add field for the podcast title and description
+# - add field for the script
+# -----------------------------------------------------------------------------
+
 # -----------------------------------------------------------------------------
 # Get default podcast materials, from Daily papers and one download
 # -----------------------------------------------------------------------------
@@ -60,6 +69,34 @@ def generate_podcast_script(subject: str, steering_question: str | None = None)
     podcast_text = full_text[dialogue_start_index:]
     return podcast_text
 
+def generate_headline_and_description(subject: str, steering_question: str | None = None) -> tuple[str, str]:
+    """Ask the LLM for a headline and a short description for the podcast episode."""
+    prompt = f"""You are a world-class podcast producer. Given the following paper or topic, generate:
+1. A catchy, informative headline for a podcast episode about it (max 15 words).
+2. A short, engaging description (2-3 sentences, max 60 words) that summarizes what listeners will learn or why the topic is exciting.
+
+Here is the topic:
+{subject[:10000]}
+"""
+    messages = [
+        {"role": "system", "content": "You are a world-class podcast producer."},
+        {"role": "user", "content": prompt},
+    ]
+    response = client.chat_completion(
+        messages,
+        max_tokens=512,
+    )
+    full_text = response.choices[0].message.content.strip()
+    # Try to split headline and description
+    lines = [l.strip() for l in full_text.splitlines() if l.strip()]
+    if len(lines) >= 2:
+        headline = lines[0]
+        description = " ".join(lines[1:])
+    else:
+        headline = full_text[:80]
+        description = full_text
+    return headline, description
+
 # -----------------------------------------------------------------------------
 # Kokoro TTS
 # -----------------------------------------------------------------------------
@@ -109,9 +146,32 @@ def generate_podcast(topic: str):
             t0 = time.time()
             ref_s = pipeline_voice[len(ps) - 1]
             audio_numpy = kmodel(ps, ref_s, speed).numpy()
-            yield (sr, audio_numpy)
+            
+            # Convert numpy array to MP3
+            with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as temp_wav:
+                sf.write(temp_wav.name, audio_numpy, sr)
+                temp_wav_path = temp_wav.name
+                
+            # Use pydub to convert WAV to MP3
+            audio_segment = AudioSegment.from_wav(temp_wav_path)
+            with tempfile.NamedTemporaryFile(suffix='.mp3', delete=False) as temp_mp3:
+                audio_segment.export(temp_mp3.name, format="mp3")
+                temp_mp3_path = temp_mp3.name
+            
+            # Read the MP3 data
+            with open(temp_mp3_path, 'rb') as mp3_file:
+                mp3_data = mp3_file.read()
+            
+            # Clean up temporary files
+            os.unlink(temp_wav_path)
+            os.unlink(temp_mp3_path)
+            
+            # Yield MP3 data instead of numpy array
+            yield (sr, mp3_data)
+            
             t1 = time.time()
-            print(f"PROCESSED '{utterance}' in {int(t1-t0)} seconds. {audio_numpy.shape}")
+            print(f"PROCESSED '{utterance}' in {int(t1-t0)} seconds. MP3 conversion completed.")
+
 
 EXAMPLES = [
     ["https://huggingface.co/blog/inference-providers-cohere", None, "How does using this compare with other inference solutions?"],
@@ -132,7 +192,7 @@ Based on [Kokoro TTS](https://huggingface.co/hexgrad/Kokoro-82M) and [Llama-3.3-
     outputs=[
         gr.Audio(
             label="Listen to your podcast! 🔊",
-            format="wav",
+            format="mp3",
             streaming=True,
         ),
     ],

requirements.txt

@@ -4,4 +4,6 @@ transformers
 PyMuPDF
 soundfile
 numpy
-requests
+requests
+pydub
+ffmpeg-python

rss.xml

@@ -26,8 +26,8 @@
 
     <!-- Example Episode -->
     <item>
-      <title>Ep 1 – “Title of the Most Upvoted Paper”</title>
-      <description>Today’s top paper on Hugging Face is: “XYZ.” Listen to an AI discuss the ideas, findings, and questions it raises. (Warning: AI may hallucinate)</description>
+      <title>Step 1x3D – From Scrap Models to Masterpieces</title>
+      <description>Today’s episode dives into Step 1x3D, a new open-source method that cleans noisy 3D data, bridges 2D–3D generation, and rivals top proprietary tools. From mesh repair to texture-perfect diffusion, it’s a major leap for 3D AI.</description>
       <pubDate>Tue, 13 May 2025 10:00:00 +0000</pubDate>
       <enclosure url="https://yourpodcastwebsite.com/audio/episode1.mp3" length="12345678" type="audio/mpeg"/>
       <guid>https://yourpodcastwebsite.com/audio/episode1.mp3</guid>

update-rss.py

@@ -0,0 +1,42 @@
+import xml.etree.ElementTree as ET
+from datetime import datetime
+import os
+from app import generate_headline_and_description
+
+def get_next_episode_number(podcast_dir="podcasts"):
+    files = [f for f in os.listdir(podcast_dir) if f.endswith(".wav")]
+    return len(files) + 1
+
+def update_rss(subject, audio_url, audio_length, rss_path="rss.xml"):
+    # Generate headline and description automatically
+    title, description = generate_headline_and_description(subject)
+    
+    tree = ET.parse(rss_path)
+    root = tree.getroot()
+    channel = root.find("channel")
+    
+    # Update lastBuildDate
+    last_build_date = channel.find("lastBuildDate")
+    now_rfc2822 = datetime.utcnow().strftime("%a, %d %b %Y %H:%M:%S +0000")
+    if last_build_date is not None:
+        last_build_date.text = now_rfc2822
+    
+    # Create new item
+    item = ET.Element("item")
+    ET.SubElement(item, "title").text = title
+    ET.SubElement(item, "description").text = description
+    ET.SubElement(item, "pubDate").text = now_rfc2822
+    ET.SubElement(item, "enclosure", url=audio_url, length=str(audio_length), type="audio/mpeg")
+    ET.SubElement(item, "guid").text = audio_url
+    ET.SubElement(item, "itunes:explicit").text = "false"
+    
+    # Insert new item after lastBuildDate (i.e., as the first item)
+    # Find the first <item> and insert before it, or append if none exist
+    items = channel.findall("item")
+    if items:
+        channel.insert(list(channel).index(items[0]), item)
+    else:
+        channel.append(item)
+    
+    # Write back to file
+    tree.write(rss_path, encoding="utf-8", xml_declaration=True)