Major Update: Kokoro-82M with 54 Premium Voices

app.py

@@ -0,0 +1,182 @@
+"""
+Kokoro-82M TTS with 54 Voices
+Built on StyleTTS 2 Architecture
+"""
+
+import gradio as gr
+import numpy as np
+import scipy.io.wavfile as wavfile
+from io import BytesIO
+import requests
+import json
+
+# Voice database - 54 voices
+VOICES = {
+    "American Female": {
+        "af_heart": "Heart - Warm & Friendly",
+        "af_bella": "Bella - Elegant & Smooth",
+        "af_nicole": "Nicole - Professional",
+        "af_aoede": "Aoede - Cheerful",
+        "af_kore": "Kore - Gentle",
+        "af_sarah": "Sarah - Clear",
+        "af_nova": "Nova - Modern",
+        "af_sky": "Sky - Light",
+        "af_alloy": "Alloy - Versatile",
+        "af_jessica": "Jessica - Natural",
+        "af_river": "River - Calm"
+    },
+    "American Male": {
+        "am_michael": "Michael - Deep & Authoritative",
+        "am_fenrir": "Fenrir - Strong",
+        "am_puck": "Puck - Playful",
+        "am_echo": "Echo - Resonant",
+        "am_eric": "Eric - Professional",
+        "am_liam": "Liam - Friendly",
+        "am_onyx": "Onyx - Rich",
+        "am_adam": "Adam - Natural"
+    },
+    "British Female": {
+        "bf_emma": "Emma - Refined",
+        "bf_isabella": "Isabella - Elegant",
+        "bf_alice": "Alice - Clear",
+        "bf_lily": "Lily - Soft"
+    },
+    "British Male": {
+        "bm_george": "George - Distinguished",
+        "bm_fable": "Fable - Storyteller",
+        "bm_lewis": "Lewis - Smooth",
+        "bm_daniel": "Daniel - Professional"
+    }
+}
+
+# Flatten voice dict for dropdown
+def get_voice_list():
+    voice_list = []
+    for category, voices in VOICES.items():
+        for voice_id, desc in voices.items():
+            voice_list.append(f"{desc} ({voice_id})")
+    return voice_list
+
+def generate_speech(text, voice_dropdown, speed):
+    """Generate speech using Kokoro-82M via HF API"""
+
+    if not text.strip():
+        return None, "❌ Please enter some text"
+
+    # Extract voice_id from dropdown selection
+    voice_id = voice_dropdown.split("(")[-1].strip(")")
+
+    try:
+        # Use Hugging Face Inference API
+        API_URL = "https://api-inference.huggingface.co/models/hexgrad/Kokoro-82M"
+
+        headers = {
+            "Content-Type": "application/json"
+        }
+
+        payload = {
+            "inputs": text,
+            "parameters": {
+                "voice": voice_id,
+                "speed": speed
+            }
+        }
+
+        response = requests.post(API_URL, headers=headers, json=payload)
+
+        if response.status_code == 200:
+            # Save audio
+            audio_bytes = response.content
+
+            # Return audio for playback
+            return audio_bytes, f"✅ Generated with {voice_id} at {speed}x speed"
+        else:
+            return None, f"❌ API Error: {response.status_code}"
+
+    except Exception as e:
+        return None, f"❌ Error: {str(e)}"
+
+# Build Gradio interface
+with gr.Blocks(title="Kokoro-82M TTS - 54 Voices", theme=gr.themes.Soft()) as demo:
+
+    gr.Markdown("""
+    # 🎙️ Kokoro-82M Text-to-Speech
+
+    **82 Million Parameters • 54 Premium Voices • StyleTTS 2 Architecture**
+
+    Choose from American & British voices with unique characteristics!
+    """)
+
+    with gr.Row():
+        with gr.Column(scale=1):
+            gr.Markdown("### 🎭 Voice Selection")
+
+            voice_selector = gr.Dropdown(
+                choices=get_voice_list(),
+                value=get_voice_list()[0],
+                label="Choose Voice (54 options)",
+                interactive=True
+            )
+
+            gr.Markdown("### ⚙️ Settings")
+            speed = gr.Slider(
+                minimum=0.5,
+                maximum=2.0,
+                value=1.0,
+                step=0.05,
+                label="Speed",
+                interactive=True
+            )
+
+            gr.Markdown("""
+            ### 🌟 Voice Categories
+            - 🇺🇸 **American Female**: 11 voices
+            - 🇺🇸 **American Male**: 8 voices
+            - 🇬🇧 **British Female**: 4 voices
+            - 🇬🇧 **British Male**: 4 voices
+            """)
+
+        with gr.Column(scale=2):
+            gr.Markdown("### 📝 Text Input")
+
+            text_input = gr.Textbox(
+                lines=5,
+                placeholder="Enter your text here... Kokoro-82M supports natural prosody and emotion!",
+                value="Welcome to Kokoro-82M! Choose from 54 premium voices powered by StyleTTS 2.",
+                label="Text to synthesize"
+            )
+
+            generate_btn = gr.Button("🎤 Generate Speech", variant="primary", size="lg")
+
+            status_text = gr.Textbox(label="Status", interactive=False)
+
+            audio_output = gr.Audio(
+                label="Generated Audio",
+                type="numpy",
+                interactive=False
+            )
+
+            gr.Markdown("""
+            ### 📊 Model Information
+            - **Model**: Kokoro-82M
+            - **Architecture**: StyleTTS 2 + ISTFTNet
+            - **Parameters**: 82 Million
+            - **License**: Apache 2.0
+            - **Training**: Few hundred hours of permissive data
+            """)
+
+    # Connect event
+    generate_btn.click(
+        fn=generate_speech,
+        inputs=[text_input, voice_selector, speed],
+        outputs=[audio_output, status_text]
+    )
+
+    gr.Markdown("""
+    ---
+    **Note**: This uses Hugging Face Inference API. First generation may take 20-30 seconds for model loading.
+    Subsequent generations are faster (~2-5 seconds).
+    """)
+
+if __name__ == "__main__":
+    demo.launch()