---

license: apache-2.0 library_name: onnx pipeline_tag: text-to-speech language: [en, de, es, it, pt, ru, hi, fr] tags: [text-to-speech, onnx, int8, safetensors, pytorch, kittentts]

Kitten TTS 2 Pico

7.2M parameters per language model, 8 languages, 30 voices, 24 kHz. ONNX with int8 weights (8.0 MB; runs on CPU, no PyTorch needed), and the float32 PyTorch weights.

Usage

pip install kittenml

from kittenml import KittenTTS
import soundfile as sf

m = KittenTTS("KittenML/kitten-tts-2-pico")          # or the path of this directory
print(m.available_voices)                 # {"en": ["Eleanor", ...], "de": ["Anna", ...], ...}
for v in m.voices("de"):                  # name and description of each voice
    print(v["name"], v["description"])

audio = m.generate("Hello there, this is Kitten TTS 2 Pico.", voice="Eleanor")
audio = m.generate("Guten Morgen! Wie geht es dir?", voice="Paul", speed=1.1)    # the voice sets the language
audio = m.generate("Guten Morgen! Wie geht es dir?", language="de")          # the language's first voice (Anna)
sf.write("out.wav", audio, 24000)

m = KittenTTS("KittenML/kitten-tts-2-pico", backend="torch")                  # PyTorch, float32 weights
m = KittenTTS("KittenML/kitten-tts-2-pico", backend="torch", device="cuda")   # on a GPU
audio = m.generate("Hello there, this is Kitten TTS 2 Pico.", voice="Eleanor")

backend="onnx" (the default) runs the int8 ONNX graph with ONNX Runtime; backend="torch" runs the float32 weights in PyTorch. Only the files the selected backend needs are downloaded.

voice takes a name from the tables below (any case: "Paul" or "paul") or its description. Without voice, a language's first voice speaks.

Voices

Each voice is a first name; its voice pack is stored under the lowercase name in voices_<lang>.npz. The English voices carry the names of the same characters in KittenTTS 2.

English (en)

voice	description
Eleanor	Somber historian (female)
Edith	Elderly grandmother soft (female)
Otis	Rumbling blues musician (male)
Frank	Grizzled war veteran gravelly (male)

German (de)

voice	description
Anna	Female 1
Mia	Female 2
Lukas	Male 1
Paul	Male 2

Spanish (es)

voice	description
Sofia	Female 1
Elena	Female 2
Pablo	Male 1
Mateo	Male 2

Italian (it)

voice	description
Francesca	Female 1
Alessandro	Male 1
Giorgio	Male 2
Davide	Male 3

Portuguese (Brazil) (pt)

voice	description
Juliana	Female 1
Larissa	Female 2
Gustavo	Male 1
Caio	Male 2

Russian (ru)

voice	description
Olga	Female 1
Tatiana	Female 2
Sergei	Male 1
Alexei	Male 2

Hindi (hi)

voice	description
Priya	Female 1
Arjun	Male 1

French (fr)

voice	description
Amelie	Female 1
Lea	Female 2
Theo	Male 1
Baptiste	Male 2

License

Apache-2.0 (the models, voice packs and the bundled G2P files).