feat: add adjustable low-cut, bass, and treble controls to post-processing and revamp Gradio UI

app.py

@@ -129,15 +129,27 @@ def ensure_wav(filepath: str) -> str:
 
 import torchaudio.functional as F_audio
 
-def post_process_audio(waveform: Tensor, sr: int) -> Tensor:
+def post_process_audio(
+    waveform: Tensor,
+    sr: int,
+    low_cut_freq: float,
+    bass_gain: float,
+    treble_gain: float,
+) -> Tensor:
     """Apply professional DSP filters to make voice audio crisp and clear."""
-    # 1. High-pass filter at 80 Hz to cut muddy sub-bass rumble and AC hums
-    waveform = F_audio.highpass_biquad(waveform, sample_rate=sr, cutoff_freq=80.0)
+    # 1. High-pass filter (Low-cut) to cut sub-bass rumble and AC hums
+    if low_cut_freq > 20:
+        waveform = F_audio.highpass_biquad(waveform, sample_rate=sr, cutoff_freq=low_cut_freq)
 
-    # 2. Treble peaking equalizer at 6000 Hz (+3.0 dB) to add air, presence, and vocal crispness
-    waveform = F_audio.equalizer_biquad(waveform, sample_rate=sr, center_freq=6000.0, gain=3.0, Q=0.707)
+    # 2. Bass EQ peaking filter at 150 Hz to add warmth or reduce muddiness
+    if bass_gain != 0:
+        waveform = F_audio.equalizer_biquad(waveform, sample_rate=sr, center_freq=150.0, gain=bass_gain, Q=0.707)
 
-    # 3. Normalize peak to -1.0 dBFS (0.9 max amplitude) to maximize volume without clipping
+    # 3. Treble EQ peaking filter at 6000 Hz to add air, presence, and vocal crispness
+    if treble_gain != 0:
+        waveform = F_audio.equalizer_biquad(waveform, sample_rate=sr, center_freq=6000.0, gain=treble_gain, Q=0.707)
+
+    # 4. Normalize peak to -1.0 dBFS (0.9 max amplitude) to maximize volume without clipping
     max_val = waveform.abs().max()
     if max_val > 0:
         waveform = waveform * (0.9 / max_val)
@@ -146,19 +158,23 @@ def post_process_audio(waveform: Tensor, sr: int) -> Tensor:
 
 
 def demo_fn(
+    input_type: str,
     speech_upl: str,
     noise_type: str,
     snr: int,
     atten_lim_db: float,
     wet_dry_mix: float,
     post_process: bool,
+    low_cut_freq: float,
+    bass_gain: float,
+    treble_gain: float,
     mic_input: Optional[str] = None,
 ):
-    if mic_input:
+    if input_type == "mic":
         speech_upl = mic_input
 
     sr = config("sr", 48000, int, section="df")
-    logger.info(f"Got parameters speech_upl: {speech_upl}, noise: {noise_type}, snr: {snr}, atten_lim_db: {atten_lim_db}, wet_dry_mix: {wet_dry_mix}, post_process: {post_process}")
+    logger.info(f"Got parameters input_type: {input_type}, speech_upl: {speech_upl}, noise: {noise_type}, snr: {snr}, atten_lim_db: {atten_lim_db}, wet_dry_mix: {wet_dry_mix}, post_process: {post_process}, low_cut: {low_cut_freq}, bass: {bass_gain}, treble: {treble_gain}")
     snr = int(snr)
     noise_fn = NOISES[noise_type]
     meta = AudioMetaData(-1, -1, -1, -1, "")
@@ -204,9 +220,9 @@ def demo_fn(
     alpha = wet_dry_mix / 100.0
     enhanced = alpha * enhanced + (1 - alpha) * sample
 
-    # Apply professional post-processing (Low-cut filter + presence boost EQ)
+    # Apply professional post-processing (Low-cut filter + bass EQ + treble EQ)
     if post_process:
-        enhanced = post_process_audio(enhanced, sr)
+        enhanced = post_process_audio(enhanced, sr, low_cut_freq, bass_gain, treble_gain)
 
     lim = torch.linspace(0.0, 1.0, int(sr * 0.15)).unsqueeze(0)
     lim = torch.cat((lim, torch.ones(1, enhanced.shape[-1] - lim.shape[1])), dim=1)
@@ -357,78 +373,141 @@ def toggle(choice):
         return gr.update(visible=False, value=None), gr.update(visible=True, value=None)
 
 
-with gr.Blocks() as demo:
+theme = gr.themes.Soft(
+    primary_hue="blue",
+    secondary_hue="indigo",
+    neutral_hue="slate",
+    font=[gr.themes.GoogleFont("Inter"), "ui-sans-serif", "sans-serif"],
+)
+
+css = """
+.gradio-container {
+    max-width: 1150px !important;
+}
+.generate-btn {
+    background: linear-gradient(135deg, #2563eb 0%, #1d4ed8 100%) !important;
+    color: white !important;
+    border: none !important;
+    font-weight: 600 !important;
+    transition: all 0.2s ease-in-out !important;
+    box-shadow: 0 4px 6px -1px rgba(37, 99, 235, 0.2) !important;
+}
+.generate-btn:hover {
+    background: linear-gradient(135deg, #1d4ed8 0%, #1e40af 100%) !important;
+    transform: translateY(-1px) !important;
+    box-shadow: 0 6px 15px rgba(37, 99, 235, 0.35) !important;
+}
+"""
+
+with gr.Blocks(theme=theme, css=css) as demo:
     with gr.Row():
         gr.Markdown(
             """
-            ## DeepFilterNet2 Demo\
-
-            This demo denoises audio files using DeepFilterNet. Try it with your own voice!
+            # 🎙️ DeepFilterNet Audio Studio
+            Enhance voice recordings using **DeepFilterNet2** — an ultra-fast, real-time speech enhancement model. 
+            Adjust settings below to clean up background noise and customize vocal EQ characteristics.
             """
         )
     with gr.Row():
         with gr.Column():
+            gr.Markdown("### 📂 1. Audio Source")
             radio = gr.Radio(
-                ["mic", "file"], value="file", label="How would you like to upload your audio?"
+                ["mic", "file"], value="file", label="Input Method"
             )
-            mic_input = gr.Mic(label="Input", type="filepath", visible=False)
-            audio_file = gr.Audio(type="filepath", label="Input", visible=True)
-            inputs = [
-                audio_file,
-                gr.Dropdown(
-                    label="Add background noise",
+            mic_input = gr.Mic(label="Record Voice", type="filepath", visible=False)
+            audio_file = gr.Audio(type="filepath", label="Upload Audio File", visible=True)
+
+            gr.Markdown("### 🔊 2. Noise Simulation (Optional)")
+            with gr.Group():
+                noise_select = gr.Dropdown(
+                    label="Add Background Noise",
                     choices=list(NOISES.keys()),
                     value="None",
-                ),
-                gr.Dropdown(
-                    label="Noise Level (SNR)",
+                )
+                snr_select = gr.Dropdown(
+                    label="Simulated Noise Level (SNR)",
                     choices=["-5", "0", "10", "20"],
                     value="10",
-                ),
-                gr.Slider(
+                )
+
+            gr.Markdown("### ⚙️ 3. Denoising Settings")
+            with gr.Group():
+                atten_slider = gr.Slider(
                     minimum=6,
                     maximum=35,
                     step=1,
                     value=15,
-                    label="Max Attenuation (dB) - lower is more natural/clear, higher reduces more noise",
-                ),
-                gr.Slider(
+                    label="Max Attenuation (dB) - lower keeps natural room tone, higher reduces more noise",
+                )
+                blend_slider = gr.Slider(
                     minimum=50,
                     maximum=100,
                     step=5,
                     value=90,
-                    label="Voice Naturalness Mix (%) - 100% is fully denoised, 90% blends back some original voice",
-                ),
-                gr.Checkbox(
-                    label="Post-Process (80Hz Low-Cut & Presence Boost)",
+                    label="Voice Naturalness Mix (%) - blends original signal back to preserve transients",
+                )
+
+            with gr.Accordion("🎨 4. Advanced Vocal EQ & Presence", open=False):
+                post_process_cb = gr.Checkbox(
+                    label="Enable Post-Processing EQ & Filters",
                     value=True,
-                ),
+                )
+                low_cut_slider = gr.Slider(
+                    minimum=20,
+                    maximum=200,
+                    step=10,
+                    value=80,
+                    label="Low-Cut Filter Cutoff (Hz) - removes sub-bass rumble/AC hum",
+                )
+                bass_slider = gr.Slider(
+                    minimum=-10,
+                    maximum=10,
+                    step=1,
+                    value=0,
+                    label="Voice Bass Boost (dB) at 150Hz - adds body/warmth to thin voices",
+                )
+                treble_slider = gr.Slider(
+                    minimum=-10,
+                    maximum=10,
+                    step=1,
+                    value=3,
+                    label="Voice Treble Boost (dB) at 6kHz - adds sibilance/crispness",
+                )
+
+            inputs = [
+                radio,
+                audio_file,
+                noise_select,
+                snr_select,
+                atten_slider,
+                blend_slider,
+                post_process_cb,
+                low_cut_slider,
+                bass_slider,
+                treble_slider,
                 mic_input,
             ]
-            btn = gr.Button("Generate")
+            btn = gr.Button("Generate Denoised Audio", elem_classes="generate-btn")
         with gr.Column():
+            gr.Markdown("### 📊 Results & Visualization")
             outputs = [
-                # gr.Video(type="filepath", label="Noisy audio"),
-                gr.Audio(type="filepath", label="Noisy audio"),
-                gr.Image(label="Noisy spectrogram"),
-                # gr.Video(type="filepath", label="Enhanced audio"),
-                gr.Audio(type="filepath", label="Enhanced audio"),
-                gr.Image(label="Enhanced spectrogram"),
+                gr.Audio(type="filepath", label="Noisy Audio (Input)"),
+                gr.Image(label="Noisy Spectrogram"),
+                gr.Audio(type="filepath", label="Enhanced Audio (Output)"),
+                gr.Image(label="Enhanced Spectrogram"),
             ]
     btn.click(fn=demo_fn, inputs=inputs, outputs=outputs, api_name='denoise')
     radio.change(toggle, radio, [mic_input, audio_file])
     gr.Examples(
-        [
-            ["./samples/p232_013_clean.wav", "Kitchen", "10", 15, 90, True, None],
-            ["./samples/p232_013_clean.wav", "Cafe", "10", 15, 90, True, None],
-            ["./samples/p232_019_clean.wav", "Cafe", "10", 15, 90, True, None],
-            ["./samples/p232_019_clean.wav", "River", "10", 15, 90, True, None],
+        examples=[
+            ["./samples/p232_013_clean.wav", "Kitchen", "10"],
+            ["./samples/p232_013_clean.wav", "Cafe", "10"],
+            ["./samples/p232_019_clean.wav", "Cafe", "10"],
+            ["./samples/p232_019_clean.wav", "River", "10"],
         ],
-        fn=demo_fn,
-        inputs=inputs,
-        outputs=outputs,
-        cache_examples=True,
-    ),
+        inputs=[audio_file, noise_select, snr_select],
+        label="💡 Quick Start Demo Examples (Click to load)",
+    )
     gr.Markdown(open("usage.md").read())
 
 cleanup_tmp()