add app.py (AudioSR CPU + motyw Lektora)

app.py

@@ -0,0 +1,132 @@
+"""AudioSR — wzbogacanie krótkich próbek głosu (HF Space, CPU/DARMOWY). UI w kolorystyce
+Lektor AI (ciemny + cyan #00ffff). Pole do wrzucenia próbki → odtwarzacz + pobieranie WAV.
+
+Działa bez PRO (CPU basic). UWAGA: na darmowym CPU 1 klip ≈ ~2 min (to dyfuzja).
+Domyślnie model „basic", 24 kroki (sprawdzone do Twojego głosu — patrz sr_dataset.py)."""
+import os
+import tempfile
+import numpy as np
+import soundfile as sf
+import gradio as gr
+import torch
+from audiosr import build_model, super_resolution
+
+DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
+OUTPUT_SR = 48000
+_CACHE = {}
+
+
+def _model(model_name: str):
+    if model_name not in _CACHE:
+        _CACHE[model_name] = build_model(model_name=model_name, device=DEVICE)
+    return _CACHE[model_name]
+
+
+def enhance(audio_path, model_name, ddim_steps, guidance_scale, seed):
+    if not audio_path:
+        raise gr.Error("Najpierw wrzuć plik audio.")
+    seed = int(seed or 0)
+    if seed == 0:
+        seed = int.from_bytes(os.urandom(4), "big") % (2 ** 31)
+    wav = super_resolution(
+        _model(model_name), audio_path,
+        seed=seed, guidance_scale=float(guidance_scale),
+        ddim_steps=int(ddim_steps), latent_t_per_second=12.8,
+    )
+    wav = np.asarray(wav).squeeze()
+    if wav.ndim > 1:
+        wav = wav.T
+    # przytnij padding AudioSR (zaokrągla do 5.12 s + cisza) do długości wejścia
+    info = sf.info(audio_path)
+    exp = int(round(info.frames / info.samplerate * OUTPUT_SR))
+    if exp and wav.shape[0] > exp:
+        wav = wav[:exp]
+    # wyrównaj poziom (RMS) do oryginału — AudioSR normalizuje per-call
+    src, _ = sf.read(audio_path, dtype="float32")
+    if getattr(src, "ndim", 1) > 1:
+        src = src.mean(axis=1)
+    r_in = float(np.sqrt((src ** 2).mean())) + 1e-9
+    r_out = float(np.sqrt((wav ** 2).mean())) + 1e-9
+    wav = wav * (r_in / r_out)
+    peak = float(np.abs(wav).max()) or 1.0
+    if peak > 0.99:
+        wav = wav / peak * 0.99
+    # zapis do pliku — żeby był odtwarzacz I pobieranie czystego WAV
+    base = os.path.splitext(os.path.basename(audio_path))[0]
+    out_path = os.path.join(tempfile.gettempdir(), f"{base}_48k.wav")
+    sf.write(out_path, wav.astype(np.float32), OUTPUT_SR)
+    return out_path, gr.DownloadButton(value=out_path, interactive=True)
+
+
+# ── Motyw Lektor AI (ciemny + cyan) ──────────────────────────────────────────
+CYAN, PANEL, BG, BG2, TXT, MUTED, BORDER = (
+    "#00ffff", "#0a0a0f", "#0d0d14", "#15151e", "#cfcfe0", "#8a8aa0", "#222230")
+
+CSS = f"""
+.gradio-container, .gradio-container .dark, body {{
+  background:{PANEL} !important; color:{TXT} !important;
+  font-family:'Segoe UI', system-ui, -apple-system, sans-serif !important;
+}}
+.gradio-container {{ max-width:920px !important; margin:0 auto !important; }}
+#lk-menu {{
+  display:flex; align-items:center; gap:14px;
+  padding:14px 18px; margin:0 0 16px 0; border-radius:12px;
+  background:linear-gradient(180deg,{BG2},{BG});
+  border:1px solid {BORDER}; border-bottom:2px solid {CYAN};
+  box-shadow:0 10px 28px -16px {CYAN};
+}}
+#lk-menu .lk-title {{ font-size:20px; font-weight:700; letter-spacing:.4px; color:{TXT}; line-height:1.1; }}
+#lk-menu .lk-title b {{ color:{CYAN}; text-shadow:0 0 10px rgba(0,255,255,.45); }}
+#lk-menu .lk-sub {{ font-size:12px; color:{MUTED}; margin-top:2px; }}
+.block, .form, .gr-box, fieldset, .gr-panel, .panel {{
+  background:{BG} !important; border:1px solid {BORDER} !important; border-radius:10px !important;
+}}
+label, label span, .gr-input-label, span[data-testid="block-info"] {{ color:{MUTED} !important; }}
+input, textarea, select, .gr-text-input {{
+  background:{BG2} !important; color:{TXT} !important; border:1px solid {BORDER} !important;
+}}
+input:focus, textarea:focus, select:focus {{ border-color:{CYAN} !important; box-shadow:0 0 0 1px {CYAN} !important; }}
+input[type=range] {{ accent-color:{CYAN} !important; }}
+.gr-button-primary, button.primary, .primary {{
+  background:{CYAN} !important; color:#06141c !important; border:none !important;
+  font-weight:700 !important; letter-spacing:.6px !important;
+}}
+.gr-button-primary:hover, button.primary:hover {{ box-shadow:0 0 16px rgba(0,255,255,.55) !important; }}
+.gr-button-secondary, button.secondary {{
+  background:transparent !important; color:{CYAN} !important; border:1px solid {CYAN} !important;
+}}
+.gradio-container a {{ color:{CYAN} !important; }}
+footer {{ display:none !important; }}
+"""
+
+ICON = (f"<svg width='30' height='30' viewBox='0 0 24 24' fill='none' stroke='{CYAN}' "
+        f"stroke-width='2.2' stroke-linecap='round'>"
+        f"<line x1='4' y1='15' x2='4' y2='9'/><line x1='9' y1='19' x2='9' y2='5'/>"
+        f"<line x1='14' y1='16' x2='14' y2='8'/><line x1='19' y1='13' x2='19' y2='11'/></svg>")
+MENU = (f"<div id='lk-menu'>{ICON}<div>"
+        f"<div class='lk-title'>Audio<b>SR</b> · Lektor AI</div>"
+        f"<div class='lk-sub'>Wzbogacanie pr��bek głosu — super-rozdzielczość do 48 kHz</div>"
+        f"</div></div>")
+
+with gr.Blocks(title="AudioSR · Lektor AI", css=CSS, theme=gr.themes.Base()) as demo:
+    gr.HTML(MENU)
+    with gr.Row():
+        with gr.Column(scale=1):
+            audio = gr.Audio(sources=["upload", "microphone"], type="filepath",
+                             label="Próbka głosu — wrzuć plik (krótki klip)")
+            model = gr.Radio(["basic", "speech"], value="basic",
+                             label="Model (basic = sprawdzony do Twojego głosu)")
+            steps = gr.Slider(10, 100, value=24, step=1, label="DDIM steps (24 = jakość/czas)")
+            guid = gr.Slider(1.0, 10.0, value=3.5, step=0.5, label="Guidance scale")
+            seed = gr.Number(value=0, label="Seed (0 = losowy)")
+            btn = gr.Button("WZBOGAĆ", variant="primary")
+        with gr.Column(scale=1):
+            out = gr.Audio(type="filepath", label="Wzbogacone — 48 kHz (odtwarzacz)", interactive=False)
+            dl = gr.DownloadButton("Pobierz WAV (48 kHz)", variant="secondary", interactive=False)
+            gr.HTML(f"<div style='color:{MUTED};font-size:12px;margin-top:6px'>"
+                    f"Na darmowym CPU 1 klip ≈ 2 min (dyfuzja). Model „basic”, padding przycięty, "
+                    f"poziom wyrównany do oryginału.</div>")
+    btn.click(enhance, [audio, model, steps, guid, seed], [out, dl])
+
+if __name__ == "__main__":
+    demo.queue(max_size=8).launch()