Fix podcast playback: valid MP3 output, Gradio file route, and TTS input cap

pages/artifacts.py

@@ -60,6 +60,7 @@ def _render_artifact_content(artifact) -> str:
                 f'border-radius:10px;">'
                 f'<div style="font-size:0.82rem; color:#8090d0; margin-bottom:8px;">🎧 Podcast Audio</div>'
                 f'<audio controls style="width:100%; border-radius:6px;">'
+                f'<source src="/gradio_api/file={artifact.audio_path}" type="audio/mpeg">'
                 f'<source src="/file={artifact.audio_path}" type="audio/mpeg">'
                 f'Your browser does not support the audio element.'
                 f'</audio>'

services/podcast_service.py

@@ -104,7 +104,7 @@ def _tts_request(text: str, token: str) -> bytes | None:
 
 
 def _synthesize_audio(lines: list[tuple[str, str]], output_path: Path) -> bool:
-    """Synthesize audio via OpenAI TTS. Handles 4096 char limit by chunking."""
+    """Synthesize audio via OpenAI TTS and save a valid MP3 file."""
     try:
         from openai import OpenAI
 
@@ -115,26 +115,26 @@ def _synthesize_audio(lines: list[tuple[str, str]], output_path: Path) -> bool:
         full_text = "\n".join([f"{speaker}: {text}" for speaker, text in lines])
         
         logger.info("TTS: Generating audio for %d lines (%d chars)", len(lines), len(full_text))
-        
-        # OpenAI TTS has 4096 character limit, so chunk the text
-        MAX_CHARS = 4000  # Leave buffer for safety
-        chunks = _chunk_text(full_text, MAX_CHARS)
-        logger.info("Split into %d chunks for TTS", len(chunks))
-        
-        # Generate audio for each chunk and collect them
-        audio_data = b""
-        for i, chunk in enumerate(chunks):
-            logger.info("TTS chunk %d/%d (%d chars)", i + 1, len(chunks), len(chunk))
-            response = client.audio.speech.create(
-                model="tts-1",
-                voice="alloy",  # neutral voice
-                input=chunk,
+
+        max_chars = 4000
+        if len(full_text) > max_chars:
+            trimmed = _chunk_text(full_text, max_chars)[0]
+            logger.warning(
+                "TTS input exceeded limit; trimming from %d to %d chars",
+                len(full_text),
+                len(trimmed),
             )
-            audio_data += response.content
+            full_text = trimmed
         
-        # Write combined audio to file
+        # Generate a single MP3 response and write it directly.
+        response = client.audio.speech.create(
+            model="tts-1",
+            voice="alloy",
+            input=full_text,
+        )
+
         with open(str(output_path), "wb") as f:
-            f.write(audio_data)
+            f.write(response.content)
         
         file_size = os.path.getsize(str(output_path))
         logger.info("Podcast audio saved to %s (%d bytes)", output_path, file_size)