Update app.py

app.py

@@ -1,106 +1,128 @@
-import os
-import gradio as gr
-import tempfile, os
+import os, re, tempfile
 import numpy as np
 import soundfile as sf
+import gradio as gr
 from TTS.api import TTS
 
-# If Coqui shows a CPML prompt when downloading models,
-# try pre-accept via env var (safe no-op if ignored).
-os.environ.setdefault("COQUI_TOS_AGREED", "y")
-
+# -------- speed / device --------
+USE_GPU = os.environ.get("USE_GPU", "1") == "1"  # set to 1 if you switch Space to GPU (T4, A10G)
 MODEL_NAME = "tts_models/multilingual/multi-dataset/xtts_v2"
 
-# Lazy-load to avoid heavy import before Space is ready
-_tts_obj = None
+_tts = None
 def get_tts():
-    global _tts_obj
-    if _tts_obj is None:
-        _tts_obj = TTS(MODEL_NAME)
-    return _tts_obj
+    global _tts
+    if _tts is None:
+        t = TTS(MODEL_NAME)
+        try:
+            if USE_GPU:
+                t = t.to("cuda")
+        except Exception:
+            pass
+        _tts = t
+    return _tts
 
 LANGS = [
-    ("English", "en"), ("Urdu", "ur"), ("Hindi", "hi"), ("Arabic", "ar"),
-    ("French", "fr"), ("German", "de"), ("Spanish", "es"), ("Italian", "it"),
-    ("Portuguese", "pt"), ("Turkish", "tr"),
+    ("English","en"),("Urdu","ur"),("Hindi","hi"),("Arabic","ar"),
+    ("French","fr"),("German","de"),("Spanish","es"),("Italian","it"),
+    ("Portuguese","pt"),("Turkish","tr"),
 ]
 
-def clean_text(text: str) -> str:
+def clean_text(text:str)->str:
     return " ".join((text or "").strip().split())
 
 def synth_to_file_safe(tts, txt, out_path, wav_path, lang, speed):
     try:
-        tts.tts_to_file(
-            text=txt, file_path=out_path,
-            speaker_wav=wav_path, language=lang, speed=speed,
-        )
+        tts.tts_to_file(text=txt, file_path=out_path,
+                        speaker_wav=wav_path, language=lang, speed=speed)
     except TypeError:
-        tts.tts_to_file(
-            text=txt, file_path=out_path,
-            speaker_wav=wav_path, language=lang,
-        )
+        tts.tts_to_file(text=txt, file_path=out_path,
+                        speaker_wav=wav_path, language=lang)
 
-def tts_clone(text, ref_audio, language_code, speed, split_sentences):
-    if ref_audio is None:
+def split_sentences(text:str):
+    parts = [s.strip() for s in re.split(r'(?<=[.!?])\s+', text) if s.strip()]
+    return parts or [text]
+
+def tts_clone(text, ref_audio, language_code, speed, split_long, progress=gr.Progress()):
+    if not ref_audio:
         raise gr.Error("Please upload a reference voice sample (10–60 seconds).")
+
     text = clean_text(text)
     if not text:
         raise gr.Error("Please enter some text.")
 
+    # hard guard for CPU: very long text can take a long time
+    if not USE_GPU and len(text) > 600:
+        raise gr.Error("Text is long for CPU. Please try ≤ 600 characters, or switch the Space to a GPU for long texts.")
+
     tts = get_tts()
     wav_path = ref_audio
 
-    chunks = [text]
-    if split_sentences:
-        import re
-        chunks = [s.strip() for s in re.split(r'(?<=[.!?])\s+', text) if s.strip()]
-
+    chunks = split_sentences(text) if split_long else [text]
     out_wavs = []
+
     with tempfile.TemporaryDirectory() as td:
         for i, chunk in enumerate(chunks, 1):
+            progress((i-1)/max(1,len(chunks)), desc=f"Generating part {i}/{len(chunks)}")
             out_path = os.path.join(td, f"part_{i}.wav")
             synth_to_file_safe(tts, chunk, out_path, wav_path, language_code, speed)
             data, sr = sf.read(out_path)
             out_wavs.append((data, sr))
 
-        if len(out_wavs) == 1:
-            final_data, sr = out_wavs[0]
-        else:
-            sr = out_wavs[0][1]
-            final_data = np.concatenate([d for d, _ in out_wavs], axis=0)
-
+        sr = out_wavs[0][1]
+        final = out_wavs[0][0] if len(out_wavs)==1 else np.concatenate([d for d,_ in out_wavs], axis=0)
         final_path = os.path.join(td, "output.wav")
-        sf.write(final_path, final_data, sr)
+        sf.write(final_path, final, sr)
         return final_path
 
-with gr.Blocks(title="TalkClone - Voice Cloning & TTS", css="#warning{border-left:4px solid #22c55e;padding-left:8px;}") as demo:
-    gr.Markdown("# TalkClone — Turn Text into Speech using a Reference Audio")
-    gr.Markdown(
-        "Upload a short, clean **reference voice** (10–60s), pick a **language**, type your **text**, and generate audio. "
-        "For best results: no music/background noise, single speaker, 16kHz+ WAV/MP3."
-    )
-    with gr.Row():
-        with gr.Column():
+# ---------------- UI ----------------
+THEME = gr.themes.Soft(
+    primary_hue="blue", neutral_hue="slate"
+).set(
+    body_background_fill="#ffffff",
+    block_background_fill="#ffffff",
+    block_border_width="1px",
+    block_border_color="#e5e7eb",
+    radius_xl="14px"
+)
+
+CUSTOM_CSS = """
+/* one-column layout */
+.container {max-width: 880px; margin: 24px auto;}
+/* hide footer (“Built with Gradio”) */
+footer { display: none !important; }
+/* hide top-right toolbar (API / Settings / etc.) */
+button[aria-label="Use via API"],
+button[aria-label="Settings"],
+a[href*="gradio.app"] { display:none !important; }
+/* tighten widgets */
+.gradio-container .wrap {gap: 8px;}
+"""
+
+with gr.Blocks(theme=THEME, css=CUSTOM_CSS, fill_height=True, title="TalkClone - Voice Cloning & TTS") as demo:
+    gr.HTML('<div class="container"><h1 style="margin:0 0 8px;font-weight:700;">TalkClone — Clone a voice & generate speech</h1>'
+            '<p style="margin:0 0 16px;color:#334155;">Upload a clean reference (10–60s), choose language, enter text, then Generate.</p></div>')
+
+    with gr.Group():
+        with gr.Column(scale=1):
             ref_audio = gr.Audio(label="Reference Voice (WAV/MP3)", type="filepath")
             language = gr.Dropdown(choices=LANGS, value="en", label="Language")
-            text = gr.Textbox(label="Text", lines=5, placeholder="Type your text here...")
+            text = gr.Textbox(label="Text", lines=6, placeholder="Type your text here…")
             speed = gr.Slider(0.7, 1.3, value=1.0, step=0.05, label="Speed")
             split = gr.Checkbox(value=True, label="Auto split long text by sentence")
-            submit = gr.Button("Generate", variant="primary")
-            gr.Markdown(
-                '<div id="warning"><strong>Consent & Safety:</strong> Only clone voices you have explicit permission to use. '
-                "Avoid public-figure impersonation and disclose AI-generated audio when required by law.</div>"
-            )
-        with gr.Column():
-            output = gr.Audio(label="Cloned Speech", type="filepath", interactive=False)
-            download = gr.File(label="Download audio")
+            generate = gr.Button("Generate", variant="primary", scale=1)
+
+    with gr.Group():
+        output = gr.Audio(label="Cloned Speech", type="filepath", interactive=False)
+        download = gr.File(label="Download audio")
 
     def run_and_return(text, ref_audio, language, speed, split):
-        out_path = tts_clone(text, ref_audio, language, speed, split)
-        return out_path, out_path
+        p = tts_clone(text, ref_audio, language, speed, split)
+        return p, p
 
-    submit.click(run_and_return, inputs=[text, ref_audio, language, speed, split],
-                 outputs=[output, download])
+    generate.click(run_and_return,
+                   inputs=[text, ref_audio, language, speed, split],
+                   outputs=[output, download])
 
 if __name__ == "__main__":
-    demo.launch()
+    # hide “Use via API”, keep errors off in production
+    demo.launch(show_api=False, show_error=False)