🎙️ AI Voice Studio

Multilingual Speech AI for Text-to-Speech, Speech-to-Text, and Audio Enhancement.

AI Voice Studio is a practical multilingual Speech AI application built with open-source models and deployed on Hugging Face Spaces. It supports Uzbek, English, and Korean and is designed as a portfolio project demonstrating model integration, inference optimization, multilingual UI/UX, and GPU/CPU-ready deployment.

👉 Try AI Voice Studio on Hugging Face Spaces

Convert text into speech with language-specific Hugging Face MMS TTS models.

• 🇺🇿 Uzbek
• 🇬🇧 English
• 🇰🇷 Korean
• Lazy model loading to reduce startup memory usage
• Uzbek Latin-to-Cyrillic preprocessing for the selected MMS checkpoint

Transcribe uploaded or recorded speech using Whisper Large V3 Turbo.

• 🇺🇿 Uzbek
• 🇬🇧 English
• 🇰🇷 Korean
• Upload or microphone input
• Language-aware transcription

A lightweight audio preprocessing pipeline for improving recorded speech:

• Noise reduction
• Silence trimming
• Peak normalization
• WAV output

The application includes a dedicated voice-cloning interface and responsible-use guidance. The current public version keeps the cloning model disabled until a suitable multilingual cloning model is integrated and tested for the target languages.

Responsible use: only use your own voice or voice samples for which you have explicit permission. Do not use voice cloning for unauthorized impersonation.

The entire interface can be switched between:

• 🇺🇿 O'zbek
• 🇬🇧 English
• 🇰🇷 한국어

Changing the interface language updates navigation, headings, labels, buttons, descriptions, developer information, and other visible UI text.

The feature-level language selectors independently provide the three supported languages for TTS, STT, and the voice-cloning interface.

Models are loaded lazily, meaning a model is loaded only when its corresponding feature is used. This helps reduce unnecessary CPU/RAM/VRAM usage in constrained deployment environments.

                         AI Voice Studio
                                │
             ┌──────────────────┼──────────────────┐
             │                  │                  │
             ▼                  ▼                  ▼
       Text → Speech      Speech → Text      Audio Enhancement
             │                  │                  │
             ▼                  ▼                  ▼
        Facebook MMS         Whisper       Librosa + Noisereduce
         UZ / EN / KO      Large V3 Turbo        + SciPy
             │                  │                  │
             └──────────────────┼──────────────────┘
                                ▼
                         Audio / Text Output

• Python 3.10
• PyTorch
• Transformers
• Gradio
• Hugging Face Spaces
• Hugging Face Spaces SDK
• Whisper
• Facebook MMS TTS
• Librosa
• SoundFile
• Noisereduce
• SciPy
• NumPy

The project is structured for Hugging Face Spaces and uses @spaces.GPU for inference functions so the same application can run with ZeroGPU-style dynamic GPU allocation when available.

The application also contains CPU-compatible inference logic and can be adapted to normal CPU or NVIDIA GPU hardware.

Hugging Face Spaces
        │
        ▼
   Gradio App
        │
   ┌────┴────┐
   │         │
 CPU path   GPU path
   │         │
   └────┬────┘
        ▼
 Speech AI Models

Ai-voice-studio/
├── app.py
├── README.md
├── requirements.txt
├── logo.png
└── portfoliyo.jpg

• app.py — Gradio application, model loading, inference, multilingual UI, and styling
• requirements.txt — Python dependencies
• logo.png — AI Voice Studio branding
• portfoliyo.jpg — developer profile image

git clone https://github.com/ikromjon-gif/Ai-voice-studio.git
cd Ai-voice-studio
pip install -r requirements.txt
python app.py

The application will start with the local Gradio interface shown in the terminal.

AI-generated speech and voice technologies can be misused. This project follows a responsible-use approach:

• Use voice cloning only with your own voice or explicit permission.
• Do not impersonate another person without authorization.
• Treat generated speech as synthetic content.
• Do not use model output as a substitute for professional or safety-critical decisions.

AI Voice Studio demonstrates practical experience in:

• Speech AI
• Text-to-Speech (TTS)
• Automatic Speech Recognition (ASR)
• Multilingual AI
• Audio preprocessing
• Transformer-based model integration
• Lazy model loading
• CPU/GPU inference
• Hugging Face deployment
• Gradio UI/UX development
• Responsible AI design

Future development may include:

• High-quality multilingual voice cloning
• Better Korean and English voice quality
• Additional Uzbek speech models
• Speaker/style controls
• Speech segmentation and timestamps
• More advanced audio restoration
• API deployment for external applications

TOJIBOEV IKROMJON MAHKHAMBOY UGLI
AI/ML Engineer · Computer Engineering · Chonnam National University

📧 ikromjonkorealife@gmail.com

Focused on practical AI applications, Speech AI, Computer Vision, and open-source machine learning models.

⭐ AI Voice Studio — Open-source Speech AI for UZ / EN / KO.