Upload app.py

app.py

@@ -173,101 +173,162 @@ SOURCE_LANGUAGES = ["English", "Chinese", "Japanese", "Korean", "German",
                     "Hindi", "Swahili", "Auto-detect"]
 
 def translate_text(client, text, source_lang, target_lang):
-    """Translate text between languages using the LLM."""
+    """Translate text between languages. Handles long texts by chunking."""
     if not client:
-        raise gr.Error("DASHSCOPE_API_KEY needed for translation. Add it in Settings > Secrets.")
+        raise gr.Error("DASHSCOPE_API_KEY needed for translation.")
     if source_lang == target_lang:
         return text
-    
-    source_hint = f" (source language: {source_lang})" if source_lang != "Auto-detect" else ""
-    response = client.chat.completions.create(
-        model=OMNI_MODEL, modalities=["text"],
-        messages=[{
-            "role": "system",
-            "content": f"Translate the following text into {target_lang}. Output ONLY the translation, nothing else.",
-        }, {
-            "role": "user",
-            "content": f"Translate this{source_hint}:\n\n{text}",
-        }],
-    )
-    translated = response.choices[0].message.content.strip()
-    print(f"[Translate] {source_lang} -> {target_lang}: {len(text)} -> {len(translated)} chars")
-    return translated
+
+    chunks = split_for_llm(text, max_chars=4000)
+    translated_parts = []
+
+    for ci, chunk in enumerate(chunks):
+        source_hint = f" (source language: {source_lang})" if source_lang != "Auto-detect" else ""
+        response = client.chat.completions.create(
+            model=OMNI_MODEL, modalities=["text"],
+            messages=[{
+                "role": "system",
+                "content": f"Translate the following text into {target_lang}. Output ONLY the translation.",
+            }, {
+                "role": "user",
+                "content": f"Translate this{source_hint}:\n\n{chunk}",
+            }],
+        )
+        translated_parts.append(response.choices[0].message.content.strip())
+        print(f"[Translate] Chunk {ci+1}/{len(chunks)} done")
+
+    result = "\n\n".join(translated_parts)
+    print(f"[Translate] {source_lang} -> {target_lang}: {len(text)} -> {len(result)} chars")
+    return result
 
 
 # ==========================================
 # EMOTION ANALYSIS
 # ==========================================
+def split_for_llm(text, max_chars=4000):
+    """Split long text into chunks at paragraph boundaries for LLM processing."""
+    if len(text) <= max_chars:
+        return [text]
+    chunks, paragraphs, current = [], re.split(r'\n\s*\n', text), ""
+    for para in paragraphs:
+        para = para.strip()
+        if not para:
+            continue
+        if len(current) + len(para) + 2 > max_chars and current:
+            chunks.append(current.strip())
+            current = para
+        else:
+            current = (current + "\n\n" + para).strip()
+    if current.strip():
+        chunks.append(current.strip())
+    return chunks if chunks else [text]
+
+
 def analyze_emotions(client, text):
-    """Split text into segments with emotion instructions."""
+    """Split text into segments with emotion instructions. Handles long texts."""
     if not client:
         return [{"text": text, "emotion": ""}]
-    try:
-        response = client.chat.completions.create(
-            model=OMNI_MODEL, modalities=["text"],
-            messages=[{
-                "role": "system",
-                "content": (
-                    "You are an audiobook director. Split text into segments where the emotional "
-                    "tone changes. For each segment, provide a specific emotion/delivery instruction.\n\n"
-                    "Output ONLY valid JSON:\n"
-                    '{"segments": [\n'
-                    '  {"text": "The lighthouse stood tall.", "emotion": "Atmospheric, steady narration"},\n'
-                    '  {"text": "She ran!", "emotion": "Urgent, breathless, rising tension"}\n'
-                    "]}\n\n"
-                    "Rules: Include ALL text. 1-4 sentences per segment. Be specific with emotions. "
-                    "No markdown. ONLY JSON."
-                ),
-            }, {"role": "user", "content": f"Direct this:\n\n{text[:6000]}"}],
-        )
-        raw = response.choices[0].message.content.strip()
-        raw = re.sub(r'^```json\s*', '', raw)
-        raw = re.sub(r'\s*```$', '', raw)
-        data = json.loads(raw)
-        segs = data.get("segments", [])
-        if segs:
-            print(f"[Emotions] {len(segs)} segments")
-            return segs
-    except Exception as e:
-        print(f"[Emotions] Failed: {e}")
-    return [{"text": text, "emotion": ""}]
+
+    chunks = split_for_llm(text, max_chars=4000)
+    all_segments = []
+
+    for ci, chunk in enumerate(chunks):
+        try:
+            response = client.chat.completions.create(
+                model=OMNI_MODEL, modalities=["text"],
+                messages=[{
+                    "role": "system",
+                    "content": (
+                        "You are an audiobook director. Split text into segments where the emotional "
+                        "tone changes. For each segment, provide a specific emotion/delivery instruction.\n\n"
+                        "Output ONLY valid JSON:\n"
+                        '{"segments": [\n'
+                        '  {"text": "The lighthouse stood tall.", "emotion": "Atmospheric, steady narration"},\n'
+                        '  {"text": "She ran!", "emotion": "Urgent, breathless, rising tension"}\n'
+                        "]}\n\n"
+                        "Rules: Include ALL text. 1-4 sentences per segment. Be specific with emotions. "
+                        "No markdown. ONLY JSON."
+                    ),
+                }, {"role": "user", "content": f"Direct this:\n\n{chunk}"}],
+            )
+            raw = response.choices[0].message.content.strip()
+            raw = re.sub(r'^```json\s*', '', raw)
+            raw = re.sub(r'\s*```$', '', raw)
+            data = json.loads(raw)
+            segs = data.get("segments", [])
+            if segs:
+                all_segments.extend(segs)
+            else:
+                all_segments.append({"text": chunk, "emotion": ""})
+            print(f"[Emotions] Chunk {ci+1}/{len(chunks)}: {len(segs)} segments")
+        except Exception as e:
+            print(f"[Emotions] Chunk {ci+1} failed: {e}")
+            all_segments.append({"text": chunk, "emotion": ""})
+
+    print(f"[Emotions] Total: {len(all_segments)} segments from {len(chunks)} chunks")
+    return all_segments if all_segments else [{"text": text, "emotion": ""}]
 
 
 def detect_characters_and_emotions(client, text):
-    """Detect characters + emotions for multi-speaker mode."""
-    response = client.chat.completions.create(
-        model=OMNI_MODEL, modalities=["text"],
-        messages=[{
-            "role": "system",
-            "content": (
-                "You are an audiobook director. Analyze this story:\n"
-                "1. Identify all characters with genders\n"
-                "2. Split into segments by speaker\n"
-                "3. Add emotion instructions per segment\n\n"
-                "Output ONLY valid JSON:\n"
-                '{"characters": [{"name": "Narrator", "gender": "neutral"}, '
-                '{"name": "Elena", "gender": "female"}],\n'
-                '"segments": [{"speaker": "Narrator", "text": "...", "emotion": "Calm narration"}, '
-                '{"speaker": "Elena", "text": "...", "emotion": "Wistful, dreamy"}]}\n\n'
-                "Rules: Narrator handles non-dialogue. Include ALL text. "
-                "Be specific with emotions. No markdown. ONLY JSON."
-            ),
-        }, {"role": "user", "content": f"Direct this story:\n\n{text[:8000]}"}],
-    )
-    raw = response.choices[0].message.content.strip()
-    raw = re.sub(r'^```json\s*', '', raw)
-    raw = re.sub(r'\s*```$', '', raw)
-    try:
-        data = json.loads(raw)
-        return data.get("characters", []), data.get("segments", [])
-    except Exception:
-        return [{"name": "Narrator", "gender": "neutral"}], [{"speaker": "Narrator", "text": text, "emotion": ""}]
+    """Detect characters + emotions for multi-speaker mode. Handles long texts."""
+    chunks = split_for_llm(text, max_chars=5000)
+    all_characters = {}
+    all_segments = []
+
+    for ci, chunk in enumerate(chunks):
+        try:
+            response = client.chat.completions.create(
+                model=OMNI_MODEL, modalities=["text"],
+                messages=[{
+                    "role": "system",
+                    "content": (
+                        "You are an audiobook director. Analyze this story:\n"
+                        "1. Identify all characters with genders\n"
+                        "2. Split into segments by speaker\n"
+                        "3. Add emotion instructions per segment\n\n"
+                        "Output ONLY valid JSON:\n"
+                        '{"characters": [{"name": "Narrator", "gender": "neutral"}, '
+                        '{"name": "Elena", "gender": "female"}],\n'
+                        '"segments": [{"speaker": "Narrator", "text": "...", "emotion": "Calm narration"}, '
+                        '{"speaker": "Elena", "text": "...", "emotion": "Wistful, dreamy"}]}\n\n'
+                        "Rules: Narrator handles non-dialogue. Include ALL text. "
+                        "Be specific with emotions. No markdown. ONLY JSON."
+                    ),
+                }, {"role": "user", "content": f"Direct this story:\n\n{chunk}"}],
+            )
+            raw = response.choices[0].message.content.strip()
+            raw = re.sub(r'^```json\s*', '', raw)
+            raw = re.sub(r'\s*```$', '', raw)
+            data = json.loads(raw)
+
+            # Merge characters (keep unique by name)
+            for c in data.get("characters", []):
+                name = c.get("name", "Narrator")
+                if name not in all_characters:
+                    all_characters[name] = c
+
+            all_segments.extend(data.get("segments", []))
+            print(f"[MultiSpeaker] Chunk {ci+1}/{len(chunks)}: {len(data.get('segments', []))} segments")
+        except Exception as e:
+            print(f"[MultiSpeaker] Chunk {ci+1} failed: {e}")
+            all_segments.append({"speaker": "Narrator", "text": chunk, "emotion": ""})
+
+    # Ensure Narrator exists
+    if "Narrator" not in all_characters:
+        all_characters["Narrator"] = {"name": "Narrator", "gender": "neutral"}
+
+    characters = list(all_characters.values())
+    # Put Narrator first
+    characters.sort(key=lambda c: 0 if c["name"] == "Narrator" else 1)
+
+    print(f"[MultiSpeaker] Total: {len(characters)} characters, {len(all_segments)} segments")
+    return characters, all_segments
 
 
 # ==========================================
 # SINGLE SPEAKER: Generate with emotions
 # ==========================================
-@spaces.GPU
+@spaces.GPU(duration=600)
 def generate_single_speaker(text_input, file_input, source_lang, target_lang, speaker_label,
                              use_clone, clone_audio, clone_transcript,
                              progress=gr.Progress()):
@@ -385,7 +446,7 @@ def generate_single_speaker(text_input, file_input, source_lang, target_lang, sp
 # ==========================================
 # MULTI-SPEAKER: Generate with character voices
 # ==========================================
-@spaces.GPU
+@spaces.GPU(duration=600)
 def generate_multi_speaker(text_input, file_input, source_lang, target_lang, progress=gr.Progress()):
     resolved = resolve_text(text_input, file_input)
     if len(resolved) < 30: