freococo/F5-Myanmar-TTS

🇲🇲 F5-TTS Burmese v2 (1,025,604 Updates Foundation Model)

The official v2 release of the F5-TTS (Flow-Matching Diffusion Transformer) Burmese speech foundation model, trained on 794.5+ hours of Burmese speech across 36 full epochs (1,025,604 updates).

"They are fighting for people's freedom. I am fighting for the language's freedom.
I just want to preserve their beautiful, lovely, and brave voices embedded in AI to last forever — marking the first time in history for a massive Burmese open-source TTS foundation model."

---

🌟 What's New in v2

• Over 1 Million Updates: Pushed training from 569,780 steps (v1) to 1,025,604 updates (36 epochs) for significantly tighter phonetic alignment and tone clarity.
• 50% Smaller Model (FP16): Pruned weights down from 1.35 GB to 674 MB for 2x faster downloads and lower GPU VRAM consumption.
• Production-Grade Zero-Shot Voice Cloning: Natural conversational cadence, improved tone handling on conjuncts and Pāḷi loanwords, and zero cut-offs.
• Raw Checkpoints Available: Intermediate checkpoints (model_1020000.pt, model_1025000.pt, and model_last.pt) are included under checkpoints/ for developers wanting to continue fine-tuning.

---

⚡ Quickstart: Python Package (f5-myanmar-tts)

The easiest way to use this model is with the official PyPI package:

pip install --upgrade f5-myanmar-tts

Python (Just 3 lines):

from f5_myanmar_tts import MyanmarTTS

# Auto-downloads lightweight v2 FP16 model (~674MB) on first run
tts = MyanmarTTS()

# Generate Burmese speech
tts.speak(
    "လူသားတွေ အားလုံးကို အယုတ်အလတ်အမြတ်မရွေး ချစ်ခင်လေးစားပါ", 
    output_file="speech.wav"
)

Voice Cloning (Clone any voice in 3–5 seconds):

tts.speak(
    text="ဒါကြောင့် အုံ့မှိုင်းနေတဲ့ ကောင်းကင်ကို အမိန့်ပေးပြီး မိုးကောင်းကင် တံခါးတွေကို ဖွင့်လိုက်တယ်",
    ref_audio="my_voice.wav",
    ref_text="အပြင် မသွားရလို့ စိတ်ညစ်နေပါတယ် မိုးတွေ တအား ရွာနေပါတယ်",
    output_file="cloned_speech.wav"
)

---

📊 Model Specifications

Parameter	Specification
Architecture	Diffusion Transformer (DiT Base)
Parameters	337,138,310 (~337M)
Layers / Heads / Dim	22 layers, 16 heads, dim=1024, text_dim=512
Training Steps	36 Epochs (1,025,604 updates)
Training Audio	794.5+ Hours Burmese Speech
Sampling Rate	24,000 Hz
Vocoder	Vocos (24kHz Mel)
Vocabulary	2,626 Burmese & Pāḷi Unicode Tokens
Format	Safetensors (674 MB FP16 Pruned EMA weights)
License	Creative Commons Attribution-NonCommercial 4.0 (CC BY-NC 4.0)

---

🕊️ Dedication & Acknowledgements

1. GEMINI AI (Google): Co-engineering partner through every line of code, architecture debugging, memory optimizations, and fine-tuning pipelines.
2. F5-TTS Research Team: Yushen Chen and the creators of F5-TTS for open-sourcing this world-class flow-matching speech architecture.
3. The Brave Voices of Freedom: National Unity Government (NUG), PVTV broadcasters, independent journalists, and creators whose voices form the backbone of this heritage preservation project.

📜 License

Released under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license. Dedicated to free public research, language preservation, education, and open-source innovation.