voice-pro

🎙️ An AI-powered web application for speech recognition, translation, and dubbing

Voice-Pro is a state-of-the-art web app that transforms multimedia content creation. It integrates YouTube video downloading, voice separation, speech recognition, translation, and text-to-speech into a single, powerful tool for creators, researchers, and multilingual professionals.

• 🔊 Top-tier speech recognition: Whisper, Faster-Whisper, Whisper-Timestamped
• 🎤 Zero-shot voice cloning: F5-TTS, E2-TTS, CosyVoice (incl. Fun-CosyVoice3 — Korean and 8 more languages)
• 📢 Multilingual text-to-speech: Edge-TTS, kokoro (optional Azure TTS with your own keys — see Azure services)
• 🎥 YouTube processing & audio extraction: yt-dlp
• 🌍 Instant translation for 100+ languages: Deep-Translator (optional Azure Translator with your own keys)

A robust alternative to ElevenLabs, Voice-Pro empowers podcasters, developers, and creators with advanced voice solutions.

⚠️ Please Note

• Due to WeConnect development work, Voice-Pro development and updates are not possible for the time being.
• We have made all Voice-Pro code open source and completely free. Voice-Pro can now be freely distributed and modified by anyone.
• It works well on Windows with NVIDIA GPU. Operation on Mac and Linux has not been verified.
• Please leave your requests on the or pages.
• Troubleshooting: In most cases, issues can be resolved by deleting the installer_files folder and then running start.bat again (a clean reinstall takes only a few minutes; downloaded AI models in model/ are kept). Errors are shown in the WebUI as red toasts that stay until closed.

📰 News & History

• ⚡ Migrated the installer from Miniconda/pip to uv — dramatically faster, fully reproducible installs from a committed uv.lock. Everything stays inside installer_files/ (uv, Python, packages).
• 🐍 Upgraded runtime: Python 3.12, Torch 2.8.0+cu128 (RTX 50-series supported), Gradio 6.20.
• 🎙️ Latest ASR stack: faster-whisper 1.2.1 (large-v3-turbo, distil-large-v3.5), openai-whisper 20250625, whisper-timestamped 1.15.9. whisperX was removed (its dependency pins blocked the Gradio 6 upgrade; existing configs fall back to faster-whisper).
• 🗣️ Latest TTS stack: F5-TTS 1.1.21, kokoro 0.9.4, edge-tts 7.x, and re-vendored CosyVoice (upstream main).
• 🇰🇷 New optional TTS model: Fun-CosyVoice3-0.5B — 9 languages including Korean, selectable in the CosyVoice tab (downloads from the official HF repo on first use).
• 🧹 CUDA Toolkit and Visual Studio Build Tools are no longer required — all dependencies ship prebuilt wheels, and PyTorch bundles the CUDA runtime.
• 🛡️ Friendly to restricted / corporate PCs: no administrator rights needed — start.bat auto-downloads a portable ffmpeg if it is not installed, Whisper model downloads self-heal after interrupted/corrupted transfers, and translation automatically retries with backoff when the network rate-limits the free Google endpoint (failed lines are reported, originals kept).
• 🚨 Errors are now visible in the WebUI: every failure shows a red error toast that stays on screen until you close it (previously a 10-second warning that was easy to miss), with actionable messages for common causes (missing ffmpeg, no media registered, etc.).
• 🖥️ UI: migrated to Gradio 6 (full-width layout for all tabs, subtitle tracks shown directly in the video players).
• 🧽 uninstall.bat no longer requires administrator rights and no longer force-reboots; uninstall.bat silent runs unattended.

• We have been focusing on WeConnect development for the past few months and have not been able to manage Voice-Pro at all.
• We have decided to open source all Voice-Pro code.
• Voice-Pro is completely free and supports Windows, Mac, Linux.
• WeConnect is an application for global cultural exchange.
• Connect with people from all over the world for meaningful cultural exchanges, language learning, and international friendships.

• 🪄 Support for fine-tuned models of F5-TTS
• 🌍 Supported languages
  • English & Chinese: SWivid/F5-TTS_v1
  • Finnish: AsmoKoskinen/F5-TTS_Finnish_Model
  • French: RASPIAUDIO/F5-French-MixedSpeakers-reduced
  • Hindi: SPRINGLab/F5-Hindi-24KHz
  • Italian: alien79/F5-TTS-italian
  • Japanese: Jmica/F5TTS/JA_21999120
  • Russian: hotstone228/F5-TTS-Russian
  • Spanish: jpgallegoar/F5-Spanish

• 🔥 Removed the AI Cover feature.
• 🚀 Added support for m-bain/whisperX.

• 🐍 Built with Python 3.10.15, Torch 2.5.1+cu124, and Gradio 5.14.0.
• 🆓 Free trial supports media up to 60 seconds in length.
• 🔥 Added the AI Cover feature.
• 🎤 Introduced support for CosyVoice and kokoro.
• ⏳ Initial run downloads CozyVoice2-0.5B (9GB), which may take over an hour depending on network speed.
• 🎧 Voice samples for cloning will be continuously updated.
• 📝 Added spaCy for natural sentence-by-sentence translation and TTS.
• ☁️ Subscription version includes Microsoft Azure Translator and TTS.
• 🏪 Subscription offers unlimited usage (no 60-second limit) during the subscription period, available via .

🎥 YouTube Showcase

⭐ Key Features

1. Dubbing Studio

• YouTube video downloads & audio extraction
• Voice separation with Demucs
• Supports 100+ languages for speech recognition & translation

2. Speech Technologies

• Speech-to-Text: Whisper, Faster-Whisper, Whisper-Timestamped
• Text-to-Speech:
  • Edge-TTS: 100+ languages, 400+ voices
  • E2-TTS, F5-TTS, CosyVoice: Zero-shot cloning
  • kokoro: Ranked #2 in HuggingFace TTS Arena

3. Real-Time Translation

• Instant speech recognition
• Multilingual translation on the fly
• Customizable audio inputs

🤖 WebUI

Dubbing Studio Tab

• All-in-one hub: YouTube downloads, noise removal, subtitles, translation, & TTS
• Supports all ffmpeg-compatible formats
• Output options: WAV, FLAC, MP3
• Subtitles & recognition for 100+ languages
• TTS with speed, volume, & pitch controls

Whisper Caption Tab

• Subtitle-focused: 90+ languages
• Video-integrated subtitle display
• Word-level highlighting & denoise options

Translate Tab

• Translation for 100+ languages
• Supports subtitle files (ASS, SSA, SRT, etc.)
• Real-time voice recognition & translation

Speech Generation Tab

• Options: Edge-TTS, F5-TTS, CosyVoice, kokoro
• Celeb voice podcasts & multilingual support

🎤✨ Reference Voice

• Please request the voice you want to add on the Issues page. Issues

💻 System Requirements

• OS: Windows 10/11 (64-bit), Linux, Mac (Apple Silicon)
• GPU: NVIDIA GPU with a recent driver (>= 570 recommended; RTX 50-series supported). CUDA Toolkit installation is NOT required.
• VRAM: 4GB+ (8GB+ preferred)
• RAM: 4GB+
• Storage: 20GB+ free space
• Internet: Required

📀 Installation

Install Voice-Pro with ease using configure.bat and start.bat (use configure.sh and start.sh on Mac/Linux).

1. Get the Package

• Clone or download the latest release (Source code (zip)) from

git clone https://github.com/abus-aikorea/voice-pro.git

2. Install & Run

1. 🚀 configure.bat (optional)
   • Sets up git and ffmpeg system-wide (CUDA Toolkit / Visual Studio are no longer needed)
   • Requires administrator rights; run once
   • No admin rights? Skip it — start.bat downloads a portable ffmpeg automatically
2. 🚀 start.bat
   • Launches Voice-Pro WebUI
   • First run downloads uv + Python 3.12 and installs all dependencies from the lockfile (minutes, not hours), then downloads AI models (~10GB — this is the slow part)
   • GPU/CPU is auto-detected; override with the GPU_CHOICE environment variable (G=NVIDIA, C=CPU) or by deleting installer_files\gpu_choice.txt
   • Retry after deleting installer_files if issues arise

3. Update

• 🚀 update.bat: Re-syncs the Python environment exactly to the committed lockfile (fast)

4. Uninstall

• Run uninstall.bat or delete the folder (portable install)
• No administrator rights required; add silent for unattended removal (uninstall.bat silent)
• Only the installer_files folder is removed — your model/ and workspace/ folders are kept

🔑 Azure services (optional, .env)

By default Voice-Pro uses free services: Deep-Translator (Google's free web endpoint) for translation and Edge-TTS for speech synthesis. If you have your own Microsoft Azure subscription, you can switch both to the Azure APIs:

1. Copy .env.example to .env in the project root:

   copy .env.example .env     # Windows
   cp .env.example .env       # Mac/Linux
   

2. Fill in your Azure credentials:

   # Azure Speech Service (TTS)
   AZURE_SPEECH_KEY=your_azure_speech_key_here
   AZURE_SPEECH_REGION=eastus
   
   # Azure Translator Service
   AZURE_TRANSLATOR_KEY=your_azure_translator_key_here
   AZURE_TRANSLATOR_ENDPOINT=https://your-translator-resource.cognitiveservices.azure.com/
   AZURE_TRANSLATOR_REGION=eastus
   

3. Restart Voice-Pro. Valid keys are detected automatically at startup — translation switches to Azure Translator and the first Speech Generation tab becomes Azure-TTS.

When is this worth setting up?

• 🏢 Corporate / restricted networks: security appliances often rate-limit or block the free translate.google.com endpoint, which slows down or fails long subtitle translations. Voice-Pro retries with backoff and keeps the original text for failed lines (you will see a warning with the failure count), but Azure Translator avoids the problem entirely.
• 🗣️ Higher-quality/consistent TTS voices and higher rate limits.
• Do NOT commit .env to version control — it contains your private keys.

❓Tips & Tricks

If Browser does not run automatically

• Close the Windows-Commnad window and run start.bat again.
• Run the browser directly and enter the address displayed in the Windows-Command window (e.g. http://127.0.0.1:7870) in the address bar.

If a CUDA Out-Of-Memory error occurs

• Check the GPU memory status in Windows Task Manager - Performance tab.
• Set the Denoise level to 0 or 1. Denoise level 2 requires at least 8GB of GPU memory.
• Set Compute Type to int type. The float type has better quality, but requires more GPU memory.

How to improve the quality of subtitles?

• The quality of subtitles tends to improve with larger Whisper models, but this is not necessarily the case. large > medium > small > base > tiny
• Among compute types, float type has good performance. The int type is a model that reduces GPU usage and increases speed through model quantization. On the other hand, performance decreases.
• If you increase the denoise level, more background sounds will be removed, and only the remaining voice will be used for voice recognition. It does not always guarantee good results.

🚨 Notice

• Due to WeConnect development work, there will be no Voice-Pro updates for the time being.
• All Voice-Pro code has been made open source. It is now completely free to use.
• WeConnect is a communication platform for global cultural exchange.

⏳ SaaS Platforms for Subtitling, Translation, and TTS

The following table lists SaaS platforms supporting subtitling, translation, and text-to-speech (TTS/dubbing) functionalities. Costs are calculated for processing a 60-minute Korean video, including subtitle generation, English translation, and English dubbing, based on the latest available pricing data as of April 15, 2025.

Platform	Subtitling	Translation	TTS/Dubbing	Cost for 60-min Video (USD, Approx.)	Key Features
Maestra	✅	✅	✅	$23.70	125+ languages, real-time captions, SEO keyword extraction, 15-min free trial.
Kapwing	✅	✅	✅	$30~$40 (Pro plan, per minute)	AI subtitles, 100+ language translations, auto lip-sync dubbing, free tier.
VEED.IO	✅	✅	❌	$24~$36 (Pro plan, partial)	99.9% accurate subtitles, Instagram-optimized captions, intuitive editor.
HappyScribe	✅	✅	✅	$36~$48 (Pay-as-you-go)	120+ languages, professional proofreading, secure, meeting transcription.
Sonix	✅	✅	✅	$30~$40 (Standard plan)	54+ languages, 30-min free transcription, YouTube/Zoom integration.
Descript	✅	✅	✅	$36~$48 (Creator plan)	Text-based editing, Overdub TTS, filler word removal, 1-hour free transcription.
AppTek	✅	✅	✅	Custom pricing (Contact)	Media-focused, custom models, metadata generation, cloud-based Workbench.
Transkriptor	✅	✅	❌	$12~$18 (Pay-as-you-go)	100+ languages, YouTube link transcription, 99% accuracy, simple editor.

Cost Calculation Details

• Maestra: Premium Plan ($158/month, 1200 credits). 60-min video: 60 credits (subtitles) + 60 credits (translation) + 60 credits (dubbing) = 180 credits. Cost = (180/1200) * $158 = $23.70.
• Kapwing: Pro plan (~$24/month, limited minutes). Estimated $0.50~$0.67/min for subtitles+translation+dubbing (based on per-minute pricing trends). 60-min cost: $30~$40. Exact pricing requires confirmation.
• VEED.IO: Pro plan (~$24/month). Subtitles+translation estimated at $0.40~$0.60/min. No TTS, so partial processing. 60-min cost: $24~$36. Confirm at veed.io.
• HappyScribe: Pay-as-you-go (~$0.20/min transcription, $0.20/min translation, $0.20/min dubbing). 60-min cost: $36~$48 (assuming combined services). Confirm at happyscribe.com.
• Sonix: Standard plan (~$10/hour transcription, additional for translation/dubbing). Estimated $0.50~$0.67/min total. 60-min cost: $30~$40. Confirm at sonix.ai.
• Descript: Creator plan (~$24/month, limited hours). Estimated $0.60~$0.80/min for subtitles+translation+dubbing. 60-min cost: $36~$48. Confirm at descript.com.
• AppTek: Custom pricing for enterprise. No public per-minute rates. Contact apptek.ai for quotes.
• Transkriptor: Pay-as-you-go ($0.05~$0.10/min transcription, similar for translation). No TTS, so partial processing. 60-min cost: $12~$18. Confirm at transkriptor.com.

Notes

• Cost for 60-min Video: Costs are approximate and assume processing a 60-minute Korean video for subtitles, English translation, and English dubbing (where available). Platforms without TTS (e.g., VEED.IO, Transkriptor) reflect partial processing costs.
• Language Support: Most platforms support Korean and English. Verify specific language availability on their websites.
• Use Cases:
  • Media/Entertainment: AppTek, Maestra
  • Social Media: Kapwing, VEED.IO
  • Podcasts/Interviews: Sonix, Descript
  • E-learning/Global Content: Transkriptor, HappyScribe
• Pricing Updates: Pricing may vary due to plan changes or promotions. Check official websites for the latest details.
• For contributions or specific use case recommendations, open an issue or submit a pull request in this repository!

☕ Contributions

Hello, I'm David from the Voice-Pro team. Our team discovers the best AI technologies in the industry and provides them for anyone to use easily and conveniently. We are a small startup in Korea that has only been around for a year. We are working hard to help you and other creators produce great content.

Your ⭐⭐⭐⭐⭐ review would be greatly appreciated as it helps our business grow with you. Please help support our small team.

Thank you, ABUS Customer Service

• If you want to participate in and help us with this project, feel free to create an Issues
• If something goes wrong, please submit a Pull requests to improve this project.
• Any type of contribution is welcome.
• For inquiries related to purchases, business partnerships, technical tuning, investments, and other matters, please contact us by email. ([email protected])."
• If you like this project, please star this repository. We would greatly appreciate it. ⭐⭐⭐
• You can support Voice-Pro with a donation here:

📬 Contact

• Email: [email protected]
• Homepage (Korean): https://www.wctokyoseoul.com

🙏 Credits

• Demucs: https://github.com/facebookresearch/demucs
• yt-dlp: https://github.com/yt-dlp/yt-dlp
• gradio: https://github.com/gradio-app/gradio
• edge-TTS: https://github.com/rany2/edge-tts
• F5-TTS: https://github.com/SWivid/F5-TTS.git
• openai-whisper: https://github.com/openai/whisper
• faster-whisper: https://github.com/SYSTRAN/faster-whisper
• whisper-timestamped: https://github.com/linto-ai/whisper-timestamped
• CosyVoice: https://github.com/FunAudioLLM/CosyVoice
• kokoro: https://github.com/hexgrad/kokoro
• Deep-Translator: https://github.com/nidhaloff/deep-translator
• spaCy: https://github.com/explosion/spaCy

©️ Copyright

by ABUS