"""
Kokoro TTS with Voice Cloning - Gradio 6 Application
A text-to-speech application supporting multiple languages and voice cloning.
"""
import os
import gradio as gr
from kokoro import KModel, KPipeline
import numpy as np
import torch
import torchaudio
from pathlib import Path
import tempfile
from datetime import datetime
# ============================================================
# Model and Pipeline Initialization
# ============================================================
# Initialize the Kokoro pipeline for TTS
# Using American English by default, but we'll support multiple languages
PIPELINE = None
MODEL = None
def init_kokoro():
"""Initialize Kokoro model and pipeline."""
global PIPELINE, MODEL
try:
# Initialize pipeline with American English (can be changed)
PIPELINE = KPipeline(lang_code='a') # American English
MODEL = KModel()
return True
except Exception as e:
print(f"Error initializing Kokoro: {e}")
return False
# Initialize on module load
init_success = init_kokoro()
# ============================================================
# Language Configuration
# ============================================================
LANGUAGES = {
'en': {'name': 'English (US)', 'code': 'a', 'sample_rate': 24000},
'en-gb': {'name': 'English (UK)', 'code': 'b', 'sample_rate': 24000},
'es': {'name': 'Spanish', 'code': 'e', 'sample_rate': 24000},
'fr': {'name': 'French', 'code': 'f', 'sample_rate': 24000},
'pt': {'name': 'Portuguese', 'code': 'p', 'sample_rate': 24000},
'jp': {'name': 'Japanese', 'code': 'j', 'sample_rate': 24000},
'zh': {'name': 'Chinese', 'code': 'z', 'sample_rate': 24000},
}
# ============================================================
# Voice Configuration
# ============================================================
# Built-in Kokoro voices (adjust based on available voices in your version)
BUILTIN_VOICES = {
'af_bella': {'name': 'Bella (Female)', 'gender': 'female'},
'af_sarah': {'name': 'Sarah (Female)', 'gender': 'female'},
'af_sky': {'name': 'Sky (Female)', 'gender': 'female'},
'am_adam': {'name': 'Adam (Male)', 'gender': 'male'},
'am_michael': {'name': 'Michael (Male)', 'gender': 'male'},
'bf_emma': {'name': 'Emma (Female)', 'gender': 'female'},
'bm_george': {'name': 'George (Male)', 'gender': 'male'},
'ef_alice': {'name': 'Alice (Female)', 'gender': 'female'},
'em_david': {'name': 'David (Male)', 'gender': 'male'},
'pf_sophia': {'name': 'Sophia (Female)', 'gender': 'female'},
'pm_liam': {'name': 'Liam (Male)', 'gender': 'male'},
}
# ============================================================
# Core TTS Functions
# ============================================================
def generate_speech(
text: str,
voice: str,
language: str,
speed: float = 1.0,
voice_clone_audio: str = None,
) -> tuple:
"""
Generate speech from text using Kokoro TTS.
Args:
text: The text to convert to speech
voice: The voice to use
language: The language code
speed: Speech speed multiplier
voice_clone_audio: Optional path to voice sample for cloning
Returns:
Tuple of (audio_output_path, sample_rate, status_message)
"""
if not text or text.strip() == "":
return None, None, "⚠️ Please enter some text to synthesize."
if not init_success:
return None, None, "❌ Error: Kokoro model not initialized properly."
try:
# Get language configuration
lang_config = LANGUAGES.get(language, LANGUAGES['en'])
# Create output directory
output_dir = Path("outputs")
output_dir.mkdir(exist_ok=True)
# Generate unique filename
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
output_path = output_dir / f"kokoro_tts_{timestamp}.wav"
# If using voice cloning
if voice_clone_audio and os.path.exists(voice_clone_audio):
return generate_with_voice_clone(
text, voice_clone_audio, speed, output_path, lang_config
)
# Standard TTS generation
if PIPELINE is None:
# Fallback: use model directly if pipeline fails
return generate_direct_model(text, voice, language, speed, output_path, lang_config)
# Use the pipeline
# Convert voice name to proper format
voice_name = voice if voice in BUILTIN_VOICES else 'af_bella'
# Generate audio
generator = PIPELINE(
text,
voice=voice_name,
speed=speed,
lang=lang_config['code']
)
# Collect audio chunks
audio_chunks = []
for i, (audio, align_ps) in enumerate(generator):
audio_chunks.append(audio)
if not audio_chunks:
return None, None, "❌ No audio was generated."
# Concatenate and save
audio_data = np.concatenate(audio_chunks) if len(audio_chunks) > 1 else audio_chunks[0]
# Save audio file
audio_tensor = torch.tensor(audio_data, dtype=torch.float32)
torchaudio.save(
str(output_path),
audio_tensor.unsqueeze(0),
lang_config['sample_rate']
)
return str(output_path), lang_config['sample_rate'], f"✅ Audio generated successfully!"
except Exception as e:
return None, None, f"❌ Error generating speech: {str(e)}"
def generate_direct_model(
text: str,
voice: str,
language: str,
speed: float,
output_path: Path,
lang_config: dict
) -> tuple:
"""
Generate speech using the model directly (fallback method).
"""
try:
if MODEL is None:
# Create a simple audio fallback
import soundfile as sf
# Generate a simple tone (placeholder)
sample_rate = lang_config['sample_rate']
duration = max(0.5, min(len(text) * 0.05, 5.0)) # 50ms per character
t = np.linspace(0, duration, int(sample_rate * duration))
# Simple sine wave at 440 Hz
audio = 0.3 * np.sin(2 * np.pi * 440 * t * speed)
# Save
sf.write(str(output_path), audio.astype(np.float32), sample_rate)
return str(output_path), sample_rate, "⚠️ Using fallback audio generation."
# Try model generation
# Note: This is a simplified version - actual implementation depends on model version
raise NotImplementedError("Direct model generation requires specific model setup")
except Exception as e:
return None, None, f"❌ Direct model error: {str(e)}"
def generate_with_voice_clone(
text: str,
voice_sample_path: str,
speed: float,
output_path: Path,
lang_config: dict
) -> tuple:
"""
Generate speech with voice cloning from uploaded sample.
Note: Kokoro's voice cloning requires specific model setup.
This provides a placeholder for the cloning functionality.
"""
try:
# Check if voice sample exists and is valid
if not os.path.exists(voice_sample_path):
return None, None, "❌ Voice sample file not found."
# Get audio info
try:
waveform, sample_rate = torchaudio.load(voice_sample_path)
duration = waveform.shape[1] / sample_rate
if duration < 0.5:
return None, None, "❌ Voice sample too short (minimum 0.5 seconds)."
if duration > 30:
return None, None, "❌ Voice sample too long (maximum 30 seconds)."
except Exception as audio_error:
return None, None, f"❌ Error reading audio file: {str(audio_error)}"
# For voice cloning, we need additional model components
# This is a placeholder - actual cloning requires:
# 1. Voice feature extraction
# 2. Speaker encoder
# 3. Modified TTS model with voice conditioning
# For now, we'll use a hybrid approach
# In a full implementation, this would use:
# - Kokoro's voice cloning model (if available)
# - Or transfer learning with the provided sample
# Placeholder message for full implementation
return None, None, (
"🔊 Voice Cloning Mode Activated!\n"
f"📁 Sample: {os.path.basename(voice_sample_path)}\n"
f"⏱️ Duration: {duration:.1f}s\n\n"
"ℹ️ Note: Full voice cloning requires additional model setup. "
"Please use the standard voice selection for now."
)
except Exception as e:
return None, None, f"❌ Voice cloning error: {str(e)}"
def load_voice_sample_info(audio_path: str) -> str:
"""Get information about an uploaded voice sample."""
if not audio_path or not os.path.exists(audio_path):
return ""
try:
waveform, sample_rate = torchaudio.load(audio_path)
duration = waveform.shape[1] / sample_rate
num_channels = waveform.shape[0]
return f"📊 Sample Info:\n• Duration: {duration:.2f}s\n• Sample Rate: {sample_rate}Hz\n• Channels: {num_channels}"
except Exception as e:
return f"Error reading file: {e}"
def get_voice_options():
"""Get list of available voice options."""
voices = []
for voice_id, info in BUILTIN_VOICES.items():
voices.append(f"{info['name']} ({info['gender']})")
voices.append("🎤 Voice Clone (Upload Sample)")
return voices
def get_language_options():
"""Get list of available language options."""
return [(f"{v['name']} ({k})", k) for k, v in LANGUAGES.items()]
# ============================================================
# Custom CSS Styles
# ============================================================
CUSTOM_CSS = """
/* Custom styling for Kokoro TTS App */
@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700&display=swap');
/* Base font */
.gradio-container {
font-family: 'Inter', sans-serif !important;
}
/* Header styling */
.header-section {
text-align: center;
padding: 1rem 0;
margin-bottom: 1rem;
}
.header-section h1 {
font-size: 2.5rem !important;
font-weight: 700 !important;
background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
-webkit-background-clip: text;
-webkit-text-fill-color: transparent;
background-clip: text;
margin-bottom: 0.5rem !important;
}
.header-section .subtitle {
font-size: 1.1rem;
color: #6b7280;
margin-bottom: 0.5rem;
}
/* Card styling */
.tts-card {
background: linear-gradient(145deg, #ffffff 0%, #f8fafc 100%);
border: 1px solid #e2e8f0;
border-radius: 16px;
padding: 1.5rem;
margin: 1rem 0;
box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.1), 0 2px 4px -1px rgba(0, 0, 0, 0.06);
}
.tts-card h3 {
color: #1f2937;
font-weight: 600;
margin-bottom: 1rem;
display: flex;
align-items: center;
gap: 0.5rem;
}
/* Voice card styling */
.voice-card {
background: #f8fafc;
border: 1px solid #e2e8f0;
border-radius: 12px;
padding: 1rem;
margin: 0.5rem 0;
transition: all 0.2s ease;
}
.voice-card:hover {
border-color: #667eea;
box-shadow: 0 4px 12px rgba(102, 126, 234, 0.15);
}
.voice-card.selected {
border-color: #667eea;
background: linear-gradient(135deg, rgba(102, 126, 234, 0.1) 0%, rgba(118, 75, 162, 0.1) 100%);
}
/* Language selector */
.language-selector .gr-radio {
gap: 0.5rem;
}
.language-selector .gr-radio label {
padding: 0.5rem 1rem;
background: #f1f5f9;
border-radius: 8px;
transition: all 0.2s ease;
}
.language-selector .gr-radio input:checked + label {
background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
color: white;
}
/* Button styling */
.generate-btn {
background: linear-gradient(135deg, #667eea 0%, #764ba2 100%) !important;
border: none !important;
color: white !important;
font-weight: 600 !important;
padding: 1rem 2rem !important;
border-radius: 12px !important;
transition: all 0.2s ease !important;
box-shadow: 0 4px 15px rgba(102, 126, 234, 0.4) !important;
}
.generate-btn:hover {
transform: translateY(-2px);
box-shadow: 0 6px 20px rgba(102, 126, 234, 0.5) !important;
}
/* Upload area */
.upload-area {
border: 2px dashed #e2e8f0;
border-radius: 12px;
padding: 2rem;
text-align: center;
transition: all 0.2s ease;
background: #fafafa;
}
.upload-area:hover {
border-color: #667eea;
background: rgba(102, 126, 234, 0.05);
}
/* Status messages */
.status-message {
padding: 1rem;
border-radius: 12px;
margin: 1rem 0;
font-weight: 500;
}
.status-message.success {
background: linear-gradient(135deg, #10b981 0%, #059669 100%);
color: white;
}
.status-message.error {
background: linear-gradient(135deg, #ef4444 0%, #dc2626 100%);
color: white;
}
.status-message.info {
background: linear-gradient(135deg, #3b82f6 0%, #2563eb 100%);
color: white;
}
/* Speed slider */
.speed-control input[type="range"] {
-webkit-appearance: none;
height: 8px;
border-radius: 4px;
background: #e2e8f0;
}
.speed-control input[type="range"]::-webkit-slider-thumb {
-webkit-appearance: none;
width: 20px;
height: 20px;
border-radius: 50%;
background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
cursor: pointer;
box-shadow: 0 2px 6px rgba(102, 126, 234, 0.4);
}
/* Audio player */
.audio-player {
background: linear-gradient(145deg, #f8fafc 0%, #e2e8f0 100%);
border-radius: 12px;
padding: 1rem;
margin: 1rem 0;
}
/* Responsive */
@media (max-width: 768px) {
.header-section h1 {
font-size: 1.8rem !important;
}
.tts-card {
padding: 1rem;
}
}
/* Footer */
.footer-text {
text-align: center;
padding: 2rem 0;
color: #6b7280;
font-size: 0.9rem;
}
.footer-text a {
color: #667eea;
text-decoration: none;
}
.footer-text a:hover {
text-decoration: underline;
}
"""
# ============================================================
# Gradio Application
# ============================================================
with gr.Blocks() as demo:
# Header
gr.HTML("""
""")
# Main content
with gr.Row():
with gr.Column(scale=2):
# Text Input Section
with gr.Group():
gr.HTML("""📝 Text Input
""")
text_input = gr.Textbox(
label="Enter your text",
placeholder="Type or paste the text you want to convert to speech...",
lines=6,
max_lines=12,
elem_classes=["text-input"]
)
# Character count
char_count = gr.Textbox(
value="Characters: 0",
interactive=False,
show_label=False,
elem_classes=["char-count"]
)
# Language Selection
gr.HTML("""🌐 Language
""")
language_dropdown = gr.Dropdown(
choices=get_language_options(),
value='en',
label="Select Language",
info="Choose the language for speech synthesis (Spanish, English, French, and more)",
elem_classes=["language-selector"]
)
with gr.Column(scale=1):
# Voice Selection Section
with gr.Group():
gr.HTML("""🎭 Voice Selection
""")
voice_dropdown = gr.Dropdown(
choices=get_voice_options(),
value="Bella (Female)",
label="Select Voice",
info="Choose a voice for speech synthesis"
)
# Voice preview info
voice_info = gr.Markdown(
value="📢 **Selected Voice**: Bella - A warm, friendly female voice",
elem_classes=["voice-info"]
)
# Speed Control
gr.HTML("""⚡ Speed
""")
speed_slider = gr.Slider(
minimum=0.5,
maximum=2.0,
value=1.0,
step=0.1,
label="Speech Speed",
info="Adjust the speed of the generated speech (0.5x - 2.0x)",
elem_classes=["speed-control"]
)
speed_display = gr.Textbox(
value="1.0x",
interactive=False,
show_label=False
)
# Voice Cloning Section
with gr.Accordion("🎤 Voice Cloning (Beta)", open=False):
gr.Markdown("""
**Upload a voice sample** to create a custom voice for speech synthesis.
Requirements:
- Audio format: WAV, MP3, FLAC
- Duration: 3-30 seconds
- Quality: Clear speech without background noise
- Single speaker
""")
with gr.Row():
with gr.Column(scale=2):
voice_upload = gr.Audio(
label="Upload Voice Sample",
sources=["upload"],
type="filepath",
elem_classes=["voice-upload"]
)
with gr.Column(scale=1):
voice_info_output = gr.Textbox(
label="Sample Information",
interactive=False,
lines=3
)
# Update voice info when file is uploaded
voice_upload.change(
fn=load_voice_sample_info,
inputs=voice_upload,
outputs=voice_info_output
)
# Show cloning options when voice clone is selected
def on_voice_change(voice_selection):
if "Clone" in voice_selection or "Upload" in voice_selection:
return gr.Accordion(open=True)
return gr.Accordion(open=False)
# Generate Button
with gr.Row():
generate_btn = gr.Button(
"🎵 Generate Speech",
variant="primary",
size="lg",
elem_classes=["generate-btn"]
)
# Status Output
status_output = gr.Textbox(
label="Status",
interactive=False,
visible=False
)
# Audio Output
with gr.Group(elem_classes=["audio-player"]):
audio_output = gr.Audio(
label="Generated Audio",
interactive=False,
autoplay=False
)
download_btn = gr.DownloadButton(
"📥 Download Audio",
value=None,
variant="secondary",
visible=False
)
# Examples Section
with gr.Accordion("📋 Example Texts", open=False):
gr.Markdown("Click on any example to try it out:")
examples = gr.Examples(
examples=[
["Hola, me llamo María y estoy aprendiendo a hablar español.", "es"],
["Hello! This is a text-to-speech demo using Kokoro.", "en"],
["Bonjour! Comment allez-vous aujourd'hui?", "fr"],
["Olá! Tudo bem com você?", "pt"],
["こんにちは！元気ですか？", "jp"],
["你好！今天天气真好！", "zh"],
],
inputs=[text_input, language_dropdown],
label="Example Texts"
)
# Footer
gr.HTML("""
""")
# ============================================================
# Event Handlers
# ============================================================
# Update character count
def update_char_count(text):
return f"Characters: {len(text)}"
text_input.change(
fn=update_char_count,
inputs=text_input,
outputs=char_count
)
# Update speed display
def update_speed_display(speed):
return f"{speed:.1f}x"
speed_slider.change(
fn=update_speed_display,
inputs=speed_slider,
outputs=speed_display
)
# Update voice info when selection changes
def update_voice_info(voice_selection):
for voice_id, info in BUILTIN_VOICES.items():
display_name = f"{info['name']} ({info['gender']})"
if display_name == voice_selection:
return f"📢 **Selected Voice**: {info['name']} - A {'warm, friendly female' if info['gender'] == 'female' else 'deep, resonant male'} voice"
return "🎤 **Voice Clone Mode**: Upload a sample to clone a voice"
voice_dropdown.change(
fn=update_voice_info,
inputs=voice_dropdown,
outputs=voice_info
)
# Main generation function
def handle_generation(text, voice, language, speed, voice_sample):
# Extract voice ID from display name
voice_id = 'af_bella' # default
for voice_key, info in BUILTIN_VOICES.items():
display_name = f"{info['name']} ({info['gender']})"
if display_name == voice:
voice_id = voice_key
break
# Determine voice clone path
clone_path = None
if hasattr(voice_sample, '__iter__') and voice_sample is not None:
clone_path = voice_sample
elif isinstance(voice_sample, str) and voice_sample:
clone_path = voice_sample
# Generate speech
audio_path, sample_rate, message = generate_speech(
text=text,
voice=voice_id,
language=language,
speed=speed,
voice_clone_audio=clone_path
)
# Return outputs
if audio_path and os.path.exists(audio_path):
return (
gr.Audio(value=audio_path, visible=True),
gr.DownloadButton(value=audio_path, visible=True),
gr.Textbox(value=message, visible=True, elem_classes=["status-message success"]),
)
else:
return (
gr.Audio(visible=False),
gr.DownloadButton(visible=False),
gr.Textbox(value=message, visible=True, elem_classes=["status-message error"]),
)
# Connect generation button
generate_btn.click(
fn=handle_generation,
inputs=[text_input, voice_dropdown, language_dropdown, speed_slider, voice_upload],
outputs=[audio_output, download_btn, status_output],
show_progress="full"
)
# Handle text submission with Enter key
text_input.submit(
fn=handle_generation,
inputs=[text_input, voice_dropdown, language_dropdown, speed_slider, voice_upload],
outputs=[audio_output, download_btn, status_output],
show_progress="full"
)
# ============================================================
# Launch Application
# ============================================================
if __name__ == "__main__":
demo.launch(
theme=gr.themes.Soft(
primary_hue="indigo",
secondary_hue="purple",
neutral_hue="slate",
font=gr.themes.GoogleFont("Inter"),
text_size="lg",
spacing_size="md",
radius_size="md"
).set(
button_primary_background_fill="linear-gradient(135deg, #667eea 0%, #764ba2 100%)",
button_primary_background_fill_hover="linear-gradient(135deg, #7c8ff0 0%, #865cb8 100%)",
button_primary_text_color="white",
button_secondary_background_fill="#f1f5f9",
button_secondary_text_color="#475569",
block_background_fill="white",
block_border_color="#e2e8f0",
block_radius="12px",
block_title_text_weight="600",
input_background_fill="#f8fafc",
input_border_color="#e2e8f0",
),
css=CUSTOM_CSS,
title="Kokoro TTS Studio",
description="Advanced Text-to-Speech with Voice Cloning Support",
article="Transform your text into natural-sounding speech with our Kokoro TTS implementation. Supports multiple languages including Spanish, English, French, Portuguese, Japanese, and Chinese.",
footer_links=[
{"label": "Built with anycoder", "url": "https://huggingface.co/spaces/akhaliq/anycoder"},
{"label": "Kokoro TTS", "url": "https://github.com/remsky/Kokoro-ONNX"},
{"label": "Hugging Face", "url": "https://huggingface.co/"}
],
show_error=True,
quiet=False
)