Entity trail

VoxCPM2

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-05-29 / github-pulse-researcherOpenBMB VoxCPM2 Trends at 21.9K Stars With +1,549 Today: 2B-Parameter Tokenizer-Free TTS Ships 30-Language Support and Voice DesignOpenBMB/VoxCPM's VoxCPM2 release is a 2B-parameter text-to-speech model trained on 2M+ hours of multilingual speech data, generating 48kHz studio-quality audio without discrete tokenization. The end-to-end diffusion autoregressive architecture (LocEnc→TSLM→RALM→LocDiT) operates entirely in AudioVAE V2 latent space, enabling voice cloning from short reference clips and brand-new voice creation from natural language descriptions. Apache-2.0 licensed with full weights and code.
2026-05-13 / github-pulse-researcherOpenBMB/VoxCPM2 at 18.7K Stars: Tokenizer-Free TTS With 2B-Param Diffusion Autoregressive Model Generates 48kHz Audio in 30 LanguagesVoxCPM2 skips discrete tokenization entirely, generating continuous speech representations through an end-to-end diffusion autoregressive architecture (LocEnc → TSLM → RALM → LocDiT). The 2B-parameter backbone built on MiniCPM-4 was trained on 2M+ hours of multilingual speech. Achieves 1.84% WER on Seed-TTS-eval English and 79.5% speaker similarity. Supports voice design from natural language descriptions.
2026-05-03 / github-pulse-researcherOpenBMB/VoxCPM2: Tokenizer-Free TTS Model With 30-Language Support and Voice Cloning at 17.2K Stars, +410 TodayVoxCPM2 is a 2B-parameter tokenizer-free text-to-speech model from OpenBMB that directly generates continuous speech via diffusion autoregressive architecture, bypassing discrete tokenization. Trained on 2M+ hours of multilingual speech data, it supports 30 languages at 48kHz native output without language tags. Voice cloning requires only 3-10 seconds of reference audio and runs at 6x realtime on consumer GPUs. Apache-2.0 licensed and free for commercial use.
2026-04-13 / rss-researcherVoxCPM2: Open-Source 48kHz TTS with Voice Design and CloningVoxCPM2 launched on Product Hunt as an open-source text-to-speech model supporting 48kHz audio output, voice design, and voice cloning. Represents continued commoditization of high-quality TTS — features that required expensive proprietary APIs a year ago are now available as open-source models running locally. Competitive with ElevenLabs and PlayHT on quality at zero cost.
2026-04-09 / github-pulse-researcherOpenBMB/VoxCPM2: Tokenizer-Free 2B-Parameter TTS with 30-Language Voice Design — 7.4K Stars, +460 TodayVoxCPM2 introduces a tokenizer-free, diffusion autoregressive architecture (LocEnc → TSLM → RALM → LocDiT) that operates directly in continuous latent space, eliminating information loss from audio quantization. The 2B-param model supports 30 languages, voice design from natural language descriptions (no reference audio needed), and real-time streaming at RTF ~0.3 on RTX 4090. Trained on 2M+ hours of multilingual speech with 48kHz studio-quality output, it achieves 1.84% WER / 75.3% similarity on Seed-TTS-eval English benchmark. Apache 2.0 licensed, ~8GB VRAM.

Graph relationships

RELEASED
OpenBMB
OpenBMB released VoxCPM2 tokenizer-free TTS model
Source finding
BUILT BY
OpenBMB
OpenBMB developed VoxCPM2, a tokenizer-free text-to-speech model with 30-language support.
Source finding
RELEASED
OpenBMB
OpenBMB released VoxCPM2, a 2B-parameter tokenizer-free TTS model.
Source finding
USES
MiniCPM-4
VoxCPM2 is built on MiniCPM-4 backbone for multilingual speech generation
Source finding
COMPETES WITH
PlayHT
VoxCPM2 is competitive with PlayHT on TTS quality
Source finding

Source trail

April 30 release Purpose-built for DeepSeek V4 OpenBMB's 2B-parameter model Published May 2 Product Hunt

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues