Upload

app.py

@@ -59,45 +59,50 @@ model.eval()
    
 @spaces.GPU
 def enhance(filepath):
-    # load & resample to model SR
     wav_np, orig_sr = librosa.load(filepath, sr=None)
-    if orig_sr != SR:
-        wav_np = librosa.resample(wav_np, orig_sr, SR)
-    # to tensor + normalize
+    if orig_sr!=SR: wav_np=librosa.resample(wav_np,orig_sr,SR)
     wav = torch.from_numpy(wav_np).float().to(device)
-    norm = torch.sqrt(len(wav) / torch.sum(wav**2))
-    wav = (wav * norm).unsqueeze(0)
-    # STFT → model → ISTFT
-    amp, pha, _ = mag_phase_stft(wav, **stft_cfg, compress_factor=model_cfg["compress_factor"])
-    amp_g, pha_g = model(amp, pha)
-    out = mag_phase_istft(amp_g, pha_g, **stft_cfg, compress_factor=model_cfg["compress_factor"])
-    out = (out / norm).squeeze().cpu().numpy()
-    # resample back
-    if orig_sr != SR:
-        out = librosa.resample(out, SR, orig_sr)
-    # write to temp file
-    out_path = "enhanced.wav"
-    sf.write(out_path, out, orig_sr)
-    # compute spectrogram plot
+    norm= torch.sqrt(len(wav)/torch.sum(wav**2)); wav=(wav*norm).unsqueeze(0)
+    amp,pha,_ = mag_phase_stft(wav,**stft_cfg,compress_factor=model_cfg["compress_factor"])
+    amp_g,pha_g = model(amp,pha)
+    out = mag_phase_istft(amp_g,pha_g,**stft_cfg,compress_factor=model_cfg["compress_factor"])
+    out = (out/norm).squeeze().cpu().numpy()
+    if orig_sr!=SR: out=librosa.resample(out,SR,orig_sr)
+    sf.write("enhanced.wav", out, orig_sr)
+
     D = librosa.stft(out, n_fft=1024, hop_length=512)
-    S_db = librosa.amplitude_to_db(np.abs(D), ref=np.max)
-    fig, ax = plt.subplots(figsize=(6,3))
-    librosa.display.specshow(S_db, sr=orig_sr, hop_length=512,
-                             x_axis="time", y_axis="hz", ax=ax)
-    ax.set_title("Enhanced Spectrogram")
-    plt.colorbar(format="%+2.0f dB", ax=ax)
-    return out_path, fig
-
-demo = gr.Interface(
-    fn=enhance,
-    inputs=gr.Audio(sources=["upload","microphone"], type="filepath", label="Input Audio"),
-    outputs=[
-        gr.Audio(label="Enhanced Audio", type="filepath"),
-        gr.Plot(label="Spectrogram")
-    ],
-    title="<a href='https://github.com/RoyChao19477/SEMamba' target='_blank'>SEMamba</a>: Speech Enhancement",
-    description="Upload or record noisy speech; SEMamba enhances it and shows the spectrogram.",
-    article="<p style='text-align:center'><a href='https://arxiv.org/abs/2405.15144' target='_blank'>SEMamba: Mamba for Long-Context Speech Enhancement (SLT 2024)</a></p>"
-)
+    S = librosa.amplitude_to_db(np.abs(D), ref=np.max)
+    fig,ax=plt.subplots(figsize=(6,3))
+    librosa.display.specshow(S,sr=orig_sr,hop_length=512,x_axis="time",y_axis="hz",ax=ax)
+    ax.set_title("Enhanced Spectrogram"); plt.colorbar(format="%+2.0f dB",ax=ax)
+    return "enhanced.wav", fig
+
+# — Custom CSS —
+CSS = """
+#title {text-align:center; margin-bottom:0.2em;}
+#subtitle {text-align:center; color:#555; margin-bottom:1.5em;}
+.duplicate-button {display:block; margin:0 auto 1.5em;}
+#audio-in {border:2px dashed #aaa; border-radius:8px; padding:1em;}
+#run-btn {width:100%; margin-top:0.5em;}
+#out-audio, #spec-plot {margin-top:1em;}
+"""
+
+# — Blocks layout —
+with gr.Blocks(css=CSS, theme="soft") as demo:
+    gr.HTML("<h1 id='title'>🎧 SEMamba: Speech Enhancement</h1>")
+    gr.HTML("<p id='subtitle'>Upload or record your noisy clip, then click Enhance to boost clarity and view its spectrogram.</p>")
+    gr.DuplicateButton(value="Duplicate Space for private use", elem_classes="duplicate-button")
+
+    with gr.Row():
+        with gr.Column(scale=1):
+            audio_in  = gr.Audio(sources=["upload","microphone"], type="filepath",
+                                 label="Your Noisy Audio", elem_id="audio-in")
+            run_btn   = gr.Button("Enhance Now 🚀", variant="primary", elem_id="run-btn")
+
+        with gr.Column(scale=1):
+            audio_out = gr.Audio(type="filepath", label="Enhanced Audio", elem_id="out-audio")
+            spec_plot = gr.Plot(label="Spectrogram", elem_id="spec-plot")
+
+    run_btn.click(enhance, inputs=audio_in, outputs=[audio_out, spec_plot])
 
 demo.launch()