DviTTS

Zero-shot text-to-speech for English and six Indian languages: Hindi, Marathi, Bengali, Punjabi, Gujarati and Odia.

DviTTS

0.5B · GPU
  • Zero-shot cloning: reads new text in any voice from 3–10 s of speech and its transcript.
  • Fast: 0.37x real time on an A10G; first audio in 1.5 s (English) and 2.7 s (Hindi).
  • Light on memory: 2.2 GB of GPU memory while speaking.
  • Easy to run: pip install, a command-line tool, and sentence-by-sentence streaming.

DviTTS-Nano

110M · phone and laptop CPU
  • On-device: 351 MB of ONNX files; no GPU, no PyTorch.
  • Real time on a phone: 0.61x real time and 877 MB peak memory on a Snapdragon 8 Gen 3.
  • 18 preset voices across all seven languages.
  • Ready for apps: an Android library and demo app, plus a Python runner for any CPU.

Accuracy

Unseen speakers only. Word error (English, Whisper large-v3) and character error (Indian languages, IndicConformer), in %; lower is better.

Test setMetricDviTTSDviTTS-Nano
Seed-TTS test-enWER2.042.65
CMU ArcticWER1.452.74
LibriTTS-R testWER3.073.63
LibriSpeech-PC test-cleanWER2.05—
IndicVoices HindiCER3.533.52
FLEURS HindiCER4.525.16
FLEURS MarathiCER2.573.70
FLEURS BengaliCER2.693.54
FLEURS PunjabiCER5.235.54
FLEURS GujaratiCER3.764.54
FLEURS OdiaCER3.113.75

Voice cloning

DviTTS reads new text in the voice of a short prompt. The prompts are speakers the model never saw in training; each output is one take, not picked from several.

English

Text “It might rain this evening, so please take an umbrella when you leave the house.”

Prompt (LibriTTS-R test-clean, unseen speaker)

DviTTS output

Hindi

Text “आज शाम बारिश हो सकती है, इसलिए घर से निकलते समय छाता साथ ले जाइए।”

Prompt (FLEURS, unseen speaker)

DviTTS output

Marathi

Text “आज संध्याकाळी पाऊस पडण्याची शक्यता आहे, त्यामुळे घरातून निघताना छत्री सोबत घ्या.”

Prompt (FLEURS, unseen speaker)

DviTTS output

Bengali

Text “আজ সন্ধ্যায় বৃষ্টি হতে পারে, তাই বাড়ি থেকে বেরোনোর সময় ছাতা সঙ্গে নিন।”

Prompt (FLEURS, unseen speaker)

DviTTS output

Punjabi

Text “ਅੱਜ ਸ਼ਾਮ ਨੂੰ ਮੀਂਹ ਪੈ ਸਕਦਾ ਹੈ, ਇਸ ਲਈ ਘਰੋਂ ਨਿਕਲਦੇ ਸਮੇਂ ਛਤਰੀ ਨਾਲ ਲੈ ਜਾਓ।”

Prompt (FLEURS, unseen speaker)

DviTTS output

Gujarati

Text “આજે સાંજે વરસાદ પડી શકે છે, તેથી ઘરેથી નીકળતી વખતે છત્રી સાથે લઈ જજો.”

Prompt (FLEURS, unseen speaker)

DviTTS output

Odia

Text “ଆଜି ସନ୍ଧ୍ୟାରେ ବର୍ଷା ହୋଇପାରେ, ତେଣୁ ଘରୁ ବାହାରିବା ବେଳେ ଛତା ସାଙ୍ଗରେ ନିଅନ୍ତୁ।”

Prompt (FLEURS, unseen speaker)

DviTTS output

Hindi prompt, English text

Text “The meeting starts at 10:30 on Monday, so please be on time.”

Prompt (FLEURS, unseen speaker)

DviTTS output

Preset voices

The same sentence in each language, from both models. Voice names are labels, not the speakers’ names.

English

Text “Please call me back after the meeting ends.”

Emma female

DviTTS

DviTTS-Nano

Grace female

DviTTS

DviTTS-Nano

James male

DviTTS

DviTTS-Nano

Oliver male

DviTTS

DviTTS-Nano

Hindi

Text “कल सुबह हम स्टेशन पर मिलेंगे।”

Ananya female

DviTTS

DviTTS-Nano

Kavya female

DviTTS

DviTTS-Nano

Arjun male

DviTTS

DviTTS-Nano

Rohan male

DviTTS

DviTTS-Nano

Marathi

Text “आज संध्याकाळी पाऊस पडण्याची शक्यता आहे.”

Sneha female

DviTTS

DviTTS-Nano

Aditya male

DviTTS

DviTTS-Nano

Bengali

Text “আজ আকাশ খুব পরিষ্কার।”

Ishita female

DviTTS

DviTTS-Nano

Arnab male

DviTTS

DviTTS-Nano

Punjabi

Text “ਅਸੀਂ ਕੱਲ੍ਹ ਸਕੂਲ ਜਾਵਾਂਗੇ।”

Jasleen female

DviTTS

DviTTS-Nano

Gurpreet male

DviTTS

DviTTS-Nano

Gujarati

Text “અમે આવતીકાલે બજારમાં જઈશું.”

Hetal female

DviTTS

DviTTS-Nano

Kunal male

DviTTS

DviTTS-Nano

Odia

Text “ଆଜି ପାଗ ବହୁତ ଭଲ ଅଛି।”

Subhashree female

DviTTS

DviTTS-Nano

Debasis male

DviTTS

DviTTS-Nano