Ming-Omni TTS

Ming-Omni TTS TTS

A compact 0.5B omni-modal speech model with near-CD-quality 44.1kHz output and zero-shot voice cloning.

Jijjiirama 5,000 character limit

Daangeessii kitaaba keessan keessaa tag SSML akka itti fayyadamtan:

<speak><prosody rate="slow">Slow speech</prosody></speak>

Tag'oota mo'ellaa filatamee beekuu - cuqaasi akka tokkotti galchiin gara teekstaatti yoo ta'e:

Mo'ellaan kun kitaaba salphaa baraa, kan akka taggaa inniin linjii hin beekkamne. Akka taggaa-based emooshiniitti, mo'ellaa akka Orpheus ykn Bark.

Haalli fuula

-12 +12
0.5x 2.0x
Birrii fi Piper, VITS, MeloTTS
Oduu kee kan uumame yooka'u yooka'u. Suuraa moolaa, galchi kitaaba, fi bu'u Jijjiira.
Audion itti fufuu
0:00
Fuula Oduu Fuula Liqii dhumaa 24 sa'a keessatti
Tarree hin-ga'iin: fayyadama namaatiif. Liiziinii Kominikeeshinii irraa $5/mo
TTS.ai jaallatan? Sochii keessanitti hiika!

Fuulaa Ming-Omni TTS

Ming-omni-tts-0.5B by inclusionAI is a compact omni-modal speech model built on the BailingMM dense backbone with a patch-by-patch flow-matching audio decoder. Despite its small 500M-parameter size, it outputs 44.1kHz audio approaching CD quality and supports zero-shot voice cloning from a reference of three seconds or more. It includes built-in emotion, dialect, and even background-music control driven by JSON instructions, and is notably stable — reporting a 0.83% word error rate on Chinese benchmarks. With Apache 2.0 licensing and modest 3GB VRAM needs, it fits high-fidelity bilingual narration, emotion-controlled voice acting, and Chinese audiobook production.

Fakkeenyaaf: High-fidelity bilingual narration, emotion-controlled voice acting, Chinese audiobook content

Fuulaa Ming-Omni TTS Dhaamsa

Akkasumas

Deebi'aa
inclusionAI
Lizenz
Apache 2.0
Daandiin
free
Jijjiiramni
medium
Dhaabbilee
Ya
Afaan Oromoo
English, Chinese
Akkasumas
1000

Ming-Omni TTS Dhaamsa

Default

English
Birrii Neutral

Default (Chinese)

Chinese
Birrii Neutral

Ming-Omni TTS TTS — FAQ

It outputs 44.1kHz audio, close to CD quality — high for a model of only 0.5B parameters — thanks to its patch-by-patch flow-matching audio decoder.

Beyond voice cloning, it supports emotion, dialect, and background-music control via JSON instructions, and it is very stable, reporting a 0.83% word error rate on Chinese benchmarks.

English and Chinese, with zero-shot voice cloning from a reference clip of three seconds or longer.
← Dhaamsawwan hundaa