MiraTTS Brings Qwen2 to Bilingual Speech Synthesis
A new text-to-speech model from OpenMOSS leverages the Qwen2 architecture to generate speech in both English and Chinese.
A new open-source text-to-speech model named MiraTTS has been released by the OpenMOSS team. The model is notable for its bilingual capabilities, generating speech in both English and Chinese, and for its foundation in the powerful Qwen2 language model architecture.
MiraTTS employs a two-stage design for speech synthesis. The first stage uses the Qwen2-based model to convert input text into a spectrogram—a visual representation of sound frequencies. This spectrogram is then passed to a separate vocoder model to produce the final audio waveform.
From Research to Application
The model's architecture offers practical advantages for developers. MiraTTS supports exporting to the ONNX format, a standard that enables efficient inference across a wide range of hardware and platforms. This focus on deployment readiness, combined with its bilingual support, makes it a useful component for various non-commercial applications.
The model is available for experimentation on its Hugging Face repository, released under a Creative Commons Non-Commercial license (CC BY-NC 4.0) that encourages academic and personal use.
Sources
- Visit
YatharthS/MiraTTS
Hugging Face
More from YatharthS
All YatharthS releases →
LuxTTS Delivers Lightweight, Open-Source Speech Synthesis
The new text-to-speech model is optimized for the ONNX runtime, making it a promising option for efficient, on-device audio generation.
More in Text → Speech
All Text → Speech →
Liquid AI's LFM3-Omni-2B brings any-to-any AI to the edge
The 2-billion-parameter model combines vision, audio and speech-to-speech in a package small enough to run on device.
IBM's Granite Switch 4.1 Unifies Chat, Logic, Speech
A single endpoint now routes text conversation, decision-making, and speech synthesis across IBM's Apache-licensed Granite line.
Fish Audio's Drama 3 brings MoE to open TTS
The open-source speech model pairs a 35B mixture-of-experts backbone with voice cloning and directed performance control.
0 comments
No comments yet. Be the first to weigh in.