Upload app.py

app.py

@@ -665,7 +665,9 @@ def generate_single_speaker(text_input, file_input, source_lang, target_lang, sp
 # MULTI-SPEAKER: Generate with character voices
 # ==========================================
 @spaces.GPU(duration=600)
-def generate_multi_speaker(text_input, file_input, source_lang, target_lang, progress=gr.Progress()):
+def generate_multi_speaker(text_input, file_input, source_lang, target_lang,
+                           v0, v1, v2, v3, v4, v5, v6, v7,
+                           progress=gr.Progress()):
     resolved = resolve_text(text_input, file_input)
     if len(resolved) < 30:
         raise gr.Error("Text too short for multi-speaker.")
@@ -702,35 +704,46 @@ def generate_multi_speaker(text_input, file_input, source_lang, target_lang, pro
     char_names = [c["name"] for c in characters]
     print(f"[Multi] {len(characters)} characters: {char_names}, {len(segments)} segments")
 
-    # Assign voices by gender
+    # Assign voices — use custom assignments from dropdowns if provided
+    voice_assignments = [v0, v1, v2, v3, v4, v5, v6, v7]
     voice_map = {}
     mi, fi = 0, 0
-    for c in characters:
+    for ci, c in enumerate(characters):
         name, gender = c["name"], c.get("gender", "neutral")
-        if is_elevenlabs:
-            el_m = EL_MALE.get(target_lang, EL_MALE.get("Arabic", []))
-            el_f = EL_FEMALE.get(target_lang, EL_FEMALE.get("Arabic", []))
-            if name == "Narrator":
-                voice_map[name] = {"id": el_m[0]["id"] if el_m else "21m00Tcm4TlvDq8ikWAM", "name": el_m[0]["name"] if el_m else "Drew"}
-            elif gender == "male" and el_m:
-                voice_map[name] = {"id": el_m[mi % len(el_m)]["id"], "name": el_m[mi % len(el_m)]["name"]}
-                mi += 1
-            elif gender == "female" and el_f:
-                voice_map[name] = {"id": el_f[fi % len(el_f)]["id"], "name": el_f[fi % len(el_f)]["name"]}
-                fi += 1
+
+        # Use custom assignment if provided
+        if ci < len(voice_assignments) and voice_assignments[ci]:
+            custom_label = voice_assignments[ci]
+            if is_elevenlabs:
+                voice_name = custom_label.split("--")[0].strip()
+                voice_id = get_el_voice_id(target_lang, custom_label)
+                voice_map[name] = {"id": voice_id, "name": voice_name}
             else:
-                voice_map[name] = {"id": "21m00Tcm4TlvDq8ikWAM", "name": "Rachel"}
+                voice_map[name] = {"speaker": custom_label.split("--")[0].strip()}
         else:
-            if name == "Narrator":
-                voice_map[name] = {"speaker": "Ryan"}
-            elif gender == "male":
-                voice_map[name] = {"speaker": MALE_SPEAKERS[mi % len(MALE_SPEAKERS)]}
-                mi += 1
-            elif gender == "female":
-                voice_map[name] = {"speaker": FEMALE_SPEAKERS[fi % len(FEMALE_SPEAKERS)]}
-                fi += 1
+            # Auto-assign by gender
+            if is_elevenlabs:
+                el_m = EL_MALE.get(target_lang, EL_MALE.get("Arabic", []))
+                el_f = EL_FEMALE.get(target_lang, EL_FEMALE.get("Arabic", []))
+                if gender == "male" and el_m:
+                    voice_map[name] = {"id": el_m[mi % len(el_m)]["id"], "name": el_m[mi % len(el_m)]["name"]}
+                    mi += 1
+                elif gender == "female" and el_f:
+                    voice_map[name] = {"id": el_f[fi % len(el_f)]["id"], "name": el_f[fi % len(el_f)]["name"]}
+                    fi += 1
+                else:
+                    voice_map[name] = {"id": "21m00Tcm4TlvDq8ikWAM", "name": "Rachel"}
             else:
-                voice_map[name] = {"speaker": "Ryan"}
+                if name == "Narrator":
+                    voice_map[name] = {"speaker": "Ryan"}
+                elif gender == "male":
+                    voice_map[name] = {"speaker": MALE_SPEAKERS[mi % len(MALE_SPEAKERS)]}
+                    mi += 1
+                elif gender == "female":
+                    voice_map[name] = {"speaker": FEMALE_SPEAKERS[fi % len(FEMALE_SPEAKERS)]}
+                    fi += 1
+                else:
+                    voice_map[name] = {"speaker": "Ryan"}
     print(f"[Multi] Voice map: {voice_map}")
 
     # Generate segments
@@ -805,13 +818,28 @@ def generate_multi_speaker(text_input, file_input, source_lang, target_lang, pro
     # Assemble
     progress(0.92, desc="Assembling audiobook...")
     final_wav = os.path.join(tmp_dir, "output.wav")
+    print(f"[Multi] Concatenating {len(audio_files)} files...")
     concatenate_wavs(audio_files, final_wav)
 
+    if not os.path.exists(final_wav):
+        raise gr.Error("Assembly failed — no WAV created.")
+
+    wav_size = os.path.getsize(final_wav) / (1024 * 1024)
+    print(f"[Multi] WAV assembled: {wav_size:.1f} MB")
+
     progress(0.96, desc="Converting to MP3...")
     final_mp3 = os.path.join(OUTPUT_DIR, f"multi_{int(time.time())}.mp3")
-    subprocess.run(["ffmpeg", "-y", "-i", final_wav, "-codec:a", "libmp3lame",
+    result = subprocess.run(["ffmpeg", "-y", "-i", final_wav, "-codec:a", "libmp3lame",
                     "-b:a", "128k", "-ar", "24000", "-ac", "1", final_mp3],
-                   capture_output=True, check=True)
+                   capture_output=True, text=True)
+    if result.returncode != 0:
+        print(f"[Multi] FFmpeg error: {result.stderr[-500:]}")
+        raise gr.Error(f"MP3 conversion failed.")
+
+    if not os.path.exists(final_mp3) or os.path.getsize(final_mp3) < 100:
+        raise gr.Error("MP3 conversion produced empty file.")
+
+    print(f"[Multi] MP3 ready: {final_mp3}, {os.path.getsize(final_mp3) / (1024*1024):.1f} MB")
 
     progress(1.0, desc="Done!")
     size = os.path.getsize(final_mp3) / (1024 * 1024)
@@ -1003,6 +1031,17 @@ with gr.Blocks(title="Audiobook Generator") as demo:
                 ms_target_lang = gr.Dropdown(choices=LANGUAGES, value="English",
                                              label="Output Language",
                                              info="Language for the generated speech")
+
+                # Character detection + voice assignment
+                with gr.Accordion("Cast — Detect & Assign Voices", open=True):
+                    ms_char_btn = gr.Button("Detect Characters", variant="secondary")
+                    ms_char_info = gr.Markdown("Enter text then click 'Detect Characters' to see who's in the story.")
+                    ms_voices = []
+                    for idx in range(8):
+                        dd = gr.Dropdown(choices=SPEAKER_CHOICES, label=f"Character {idx+1}",
+                                         visible=False, allow_custom_value=True)
+                        ms_voices.append(dd)
+
                 ms_sample_btn = gr.Button("Load Sample Story", variant="secondary", size="sm")
                 ms_btn = gr.Button("Generate Multi-Speaker Audiobook", variant="primary", size="lg")
 
@@ -1012,15 +1051,70 @@ with gr.Blocks(title="Audiobook Generator") as demo:
                 with gr.Accordion("Transcript (with cast & emotions)", open=False):
                     ms_transcript = gr.Markdown()
 
-        # Section detection state
+        # States
         _ms_sections = gr.State(value=[])
+        _ms_characters = gr.State(value=[])
+
+        def detect_chars_ui(text_input, file_input, target_lang):
+            text = resolve_text(text_input, file_input)
+            client = get_llm_client()
+            if not client:
+                raise gr.Error("DASHSCOPE_API_KEY needed.")
+            characters, _ = detect_characters_and_emotions(client, text)
+            engine = get_engine(target_lang)
+
+            # Build separate male/female voice lists
+            if engine == "elevenlabs":
+                all_voices = EL_SPEAKER_CHOICES.get(target_lang, EL_SPEAKER_CHOICES.get("Arabic", []))
+                male_voices = [f"{v['name']} -- {v['desc']}" for v in ELEVENLABS_VOICES.get(target_lang, ELEVENLABS_VOICES.get("Arabic", [])) if v["gender"] == "male"]
+                female_voices = [f"{v['name']} -- {v['desc']}" for v in ELEVENLABS_VOICES.get(target_lang, ELEVENLABS_VOICES.get("Arabic", [])) if v["gender"] == "female"]
+            else:
+                all_voices = SPEAKER_CHOICES
+                male_voices = [f"{n} -- {s['desc']}" for n, s in SPEAKERS.items() if s["gender"] == "male"]
+                female_voices = [f"{n} -- {s['desc']}" for n, s in SPEAKERS.items() if s["gender"] == "female"]
+
+            info = f"**Detected {len(characters)} characters:**\n\n"
+            updates = []
+            used_male, used_female = 0, 0
+
+            for i in range(8):
+                if i < len(characters):
+                    c = characters[i]
+                    name, gender = c["name"], c.get("gender", "neutral")
+                    info += f"{i+1}. **{name}** ({gender})\n"
+
+                    # Assign unique voice per character — no duplicates
+                    if gender == "female" and female_voices:
+                        default = female_voices[used_female % len(female_voices)]
+                        used_female += 1
+                    elif gender == "male" and male_voices:
+                        default = male_voices[used_male % len(male_voices)]
+                        used_male += 1
+                    else:
+                        # Neutral/narrator — use a male voice by default
+                        default = male_voices[used_male % len(male_voices)] if male_voices else all_voices[0]
+                        used_male += 1
+
+                    updates.append(gr.update(visible=True, label=f"{name} ({gender})",
+                                             choices=all_voices, value=default))
+                else:
+                    updates.append(gr.update(visible=False))
+
+            info += "\n*Change any voice below, then click Generate.*"
+            return [info, characters] + updates
+
+        ms_char_btn.click(
+            fn=detect_chars_ui,
+            inputs=[ms_text, ms_file, ms_target_lang],
+            outputs=[ms_char_info, _ms_characters] + ms_voices,
+        )
 
         ms_detect_btn.click(fn=detect_sections_ui, inputs=[ms_file],
                             outputs=[ms_sections_info, ms_section_choice, ms_load_btn, _ms_sections])
         ms_load_btn.click(fn=load_section_ui, inputs=[ms_section_choice, _ms_sections], outputs=[ms_text])
         ms_sample_btn.click(fn=lambda: SAMPLE, outputs=ms_text)
         ms_btn.click(fn=generate_multi_speaker,
-                     inputs=[ms_text, ms_file, ms_source_lang, ms_target_lang],
+                     inputs=[ms_text, ms_file, ms_source_lang, ms_target_lang] + ms_voices,
                      outputs=[ms_audio, ms_stats, ms_transcript])
 
     gr.Markdown(
@@ -1034,4 +1128,4 @@ with gr.Blocks(title="Audiobook Generator") as demo:
     )
 
 if __name__ == "__main__":
-    demo.launch(allowed_paths=[OUTPUT_DIR])
+    demo.launch(allowed_paths=[OUTPUT_DIR, tempfile.gettempdir()], ssr_mode=False)