VoxCPM2-Burmese-Light (မြန်မာစာ + English) πŸ‡²πŸ‡² πŸ‡¬πŸ‡§

A lightweight, optimized foundation text-to-speech (TTS) model specialized strictly for Burmese (မြန်မာစာ, my-MM) and English (en), based on OpenBMB's VoxCPM2.

🌟 Key Highlights

  • Vocabulary Pruned: Stripped 28,930 Chinese/multilingual tokens, keeping 100% of Burmese UTF-8 byte tokens and English subword BPE tokens intact.
  • 65% Memory Reduction: Runs in ~3.5 GB RAM / VRAM mode with in-memory INT8 dynamic quantization (down from 10.5 GB).
  • Studio-Quality 48kHz Audio: 48,000 Hz studio sampling rate.
  • Auto Voice-Lock: Multi-sentence speaker identity consistency for audiobooks and short stories.
  • Hardware Friendly: Runs smoothly on Apple Silicon Macs (M1/M2/M3/M4), NVIDIA GPUs (≀4-6GB VRAM), and CPU-only Ubuntu Dedicated Servers.

πŸš€ Quick Usage

For complete 1-command installer, standalone WebUI, and OpenAI-compatible API server (POST /v1/audio/speech), see the official repository: https://github.com/minminthaw/voxcpm-burmese-light

Downloads last month
40
Safetensors
Model size
2B params
Tensor type
BF16
Β·
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support