Local speech recognition & synthesis
Last updated
Generate Korean speech with VoxCPM2's 30-language TTS
A 2B TTS model that generates speech in 30 languages, including Korean.
VoxCPM2 is a 2B model that turns text into 48 kHz speech in 30 languages, including Korean and Japanese. The official Web Demo supports CPU1, Apple MPS, and CUDA2; on Apple Silicon, --device auto selects MPS. The Python example below creates a first WAV through the package API3.
Generate Korean speech from text
VoxCPM2 is OpenBMB's 2B-parameter TTS4 model. Its model card lists 30 languages, including Korean, English, and Japanese, and 48 kHz output. Use it to turn text into a playable audio file. VoxCPM2 model card
Meet the CUDA and VRAM requirements first
Synthesize a first sentence and save the WAV
Synthesize a short Korean line and save it as korean.wav. Save the code below as generate.py in your working folder and run python generate.py. The output appears in the same folder.
import soundfile as sf
from voxcpm import VoxCPM
model = VoxCPM.from_pretrained("openbmb/VoxCPM2")
audio = model.generate(text="오늘 회의는 오후 두 시에 시작합니다.")
sf.write("korean.wav", audio, model.tts_model.sample_rate)python generate.py
Compare published RTF with your own task time
The card's speed measurements are tied to named hardware and runtime8 configurations, such as an RTX 4090 or Nano-VLLM. Model loading, first-audio delay, and long-text generation can feel different. Time a few scripts of fixed length on your machine and compare both audio quality and elapsed time.
Check the license and intended use
The model card and repository list Apache 2.0. Check the terms for the code and weights you plan to distribute, and review applicable policies and voice rights before publishing or commercializing generated audio. Use voice cloning9 or another person's voice only with explicit permission.
Terminology notes
CPU — The central processor that runs the operating system and general-purpose instructions; in AI workloads it also handles tasks such as preprocessing and data movement.
Back to the textCUDA — A software platform for general-purpose computing on NVIDIA GPUs.
Back to the textAPI — A defined interface that lets other code call a program’s functions.
Back to the textText-to-Speech — Technology that converts text into spoken audio with pronunciation and prosody.
Back to the textPyTorch — A software framework for building and running AI models. Check the compatible PyTorch version and hardware support along with the model.
Back to the textVRAM — Memory used by a graphics card’s GPU to store model weights and intermediate values.
Back to the textPython virtual environment — An isolated space for installing Python packages per project, helping reduce package-version conflicts.
Back to the textRuntime — The software environment that provides facilities needed while a program runs. In local AI, it can also refer to a model execution engine.
Back to the textVoice cloning — A technique that extracts speaker characteristics from a reference recording to synthesize a similar voice. Use only your own recording or one you are authorized to use.
Back to the text
Frequently asked questions
- What should you check first in this article?
- VoxCPM2 is a 2B model that turns text into 48 kHz speech in 30 languages, including Korean and Japanese. The official Web Demo supports CPU, Apple MPS, and CUDA; on Apple Silicon, `--device auto` selects MPS. The Python example below creates a first WAV through the package API.
- How much memory does it need?
- The model card targets Python 3.10+, PyTorch 2.5+, CUDA 12+, and about 8 GB of VRAM. Install packages in a virtual environment. For CUDA, choose the matching wheel from the PyTorch page.
- Where can you check the conditions and sources?
- VoxCPM2 is OpenBMB's 2B-parameter TTS model. Its model card lists 30 languages, including Korean, English, and Japanese, and 48 kHz output. Use it to turn text into a playable audio file. [VoxCPM2 model card](https://huggingface.co/openbmb/VoxCPM2)