Upload

app.py

@@ -58,55 +58,47 @@ model.eval()
 
    
 @spaces.GPU
-def enhance(path):
-    # load & resample
-    wav, orig_sr = librosa.load(path, sr=None)
-    if orig_sr != SR: wav = librosa.resample(wav, orig_sr, SR)
-    # normalize + to-tensor
-    x = torch.from_numpy(wav).float().to(device)
-    norm = torch.sqrt(len(x)/torch.sum(x**2)); x=(x*norm).unsqueeze(0)
-    # STFT → model → ISTFT
-    amp,pha,_ = mag_phase_stft(x, **stft_cfg, compress_factor=model_cfg["compress_factor"])
-    amp2,pha2 = model(amp,pha)
-    out = mag_phase_istft(amp2,pha2, **stft_cfg, compress_factor=model_cfg["compress_factor"])
-    out = (out/norm).squeeze().cpu().numpy()
-    # back to orig_sr
-    if orig_sr != SR: out = librosa.resample(out, SR, orig_sr)
-    # save
+def enhance(filepath):
+    wav_np, orig_sr = librosa.load(filepath, sr=None)
+    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)
+    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)
-    # spectrogram
+
     D = librosa.stft(out, n_fft=1024, hop_length=512)
     S = librosa.amplitude_to_db(np.abs(D), ref=np.max)
-    fig,ax=plt.subplots(figsize=(6,3))
+    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)
+    ax.set_title("Enhanced Spectrogram")
+    plt.colorbar(format="%+2.0f dB", ax=ax)
+
     return "enhanced.wav", fig
 
-CSS = """
-#title, #subtitle {text-align:center;}
-.duplicate-button {margin:1em auto; display:block;}
-#in-audio .gradio-component {border:2px dashed #888; border-radius:8px;}
-#run {width:100%; margin-top:0.5em;}
+# --- Header markdown ---
+ABOUT = """
+# SEMamba: Speech Enhancement
+Unofficial demo for [SEMamba](https://github.com/RoyChao19477/SEMamba), a Mamba-based speech enhancement model from SLT 2024.
+Upload or record a noisy audio file to enhance clarity and view the spectrogram.
 """
 
-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, hit Enhance, and see the 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="in-audio"
-            )
-            run_btn = gr.Button("Enhance Now 🚀", variant="primary", elem_id="run")
-        with gr.Column(scale=1):
-            audio_out = gr.Audio(type="filepath", label="Enhanced Audio")
-            spec_out  = gr.Plot(label="Spectrogram")
-
-    run_btn.click(enhance, inputs=audio_in, outputs=[audio_out, spec_out])
+# --- Build demo ---
+with gr.Blocks() as demo:
+    gr.Markdown(ABOUT)
+    audio_in = gr.Audio(label="Input Audio", type="filepath", interactive=True)
+    run_btn = gr.Button("Enhance", variant="primary")
+    enhanced_audio = gr.Audio(label="Enhanced Audio", type="filepath", interactive=False)
+    spectrogram = gr.Plot(label="Spectrogram")
+
+    run_btn.click(fn=enhance, inputs=audio_in, outputs=[enhanced_audio, spectrogram])
+
+    gr.Markdown("Unofficial demo by [yourname](https://github.com/RoyChao19477)")
 
 demo.launch()