Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/KittenML/KittenTTS/llms.txt

Use this file to discover all available pages before exploring further.

kitten-tts-nano is the smallest model in the KittenTTS family, designed for real-time inference on constrained hardware. It is available in two variants: standard fp32 and quantized int8.

​
Variants

VariantModel IDSizeFormat
nano (fp32)KittenML/kitten-tts-nano-0.8-fp3256 MBFull precision
nano (int8)KittenML/kitten-tts-nano-0.8-int825 MBint8 quantized
Both variants share the same 15M parameter architecture. The int8 variant uses quantization to reduce file size at the cost of minor precision loss.

​
When to use this model

  • Edge devices with limited CPU, memory, or storage
  • Real-time text-to-speech on constrained hardware
  • Embedded systems and IoT devices
  • Mobile applications where binary size matters
  • High-throughput batch workloads where inference speed is critical

​
Usage

​
nano (fp32)

from kittentts import KittenTTS
import soundfile as sf

model = KittenTTS("KittenML/kitten-tts-nano-0.8-fp32")
audio = model.generate(
    "Welcome to Kitten TTS, the ultra-lightweight text-to-speech library.",
    voice="Jasper"
)
sf.write("output.wav", audio, 24000)

​
nano (int8)

from kittentts import KittenTTS
import soundfile as sf

model = KittenTTS("KittenML/kitten-tts-nano-0.8-int8")
audio = model.generate(
    "Welcome to Kitten TTS, the ultra-lightweight text-to-speech library.",
    voice="Jasper"
)
sf.write("output.wav", audio, 24000)
Some users have reported issues with the kitten-tts-nano-0.8-int8 model. If you encounter problems, use the fp32 variant instead and open a GitHub issue so the problem can be tracked.
Models are downloaded from Hugging Face on first use and cached locally. Pass cache_dir to control where they are stored.

Build docs developers (and LLMs) love

Get started for freeTalk to us