Klein-TTS-X1.5-Pro

Klein-TTS-X1.5-Pro

Turns text into 24 kHz speech in four languages with a 13.1M parameter model and 17 voices that run faster than real time on a CPU. For developers who need a small, fast speech model.

Made by klein-ml on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Klein-TTS-X1.5-Pro is a small speech model with 13.1M parameters. It talks in four languages: English, Italian, French and Spanish, and outputs audio at 24 kHz. It comes with 17 voices that you can also blend. It is part of the Klein-TTS X1.5 family, and the card says these models can speak quicker than playback speed even on a basic CPU.

Who it suits: Developers who need a small and fast voice for apps, assistants and tools, in one of four languages.

Industry | Who uses it | How they use it | Example
Software | App developers | Add spoken output to an app without a large model | Read a status message aloud on a laptop CPU
Accessibility | Tool builders | Read text aloud in several languages | Speak a short notice in Spanish or Italian
Media | Creators | Voice short lines with a chosen voice | Generate a short narration with a named voice

If you need the highest quality, you can skip it, since the card shows the 82M Kokoro model scoring higher on two measures.

Where to use it:

• Python with Hugging Face Transformers
• A plain CPU, or a GPU for extra speed
• Windows, macOS or Linux

Use case | Who | What they get back
Text to speech | Developers | A 24 kHz mono audio file
Voice choice | Tool builders | One of 17 voices, or a blend within a language
Speed control | Creators | Speech faster or slower, trained for 0.9 to 1.1

How to set it up:

• Python with pip
• The Transformers library
• No separate speech library, because the repo ships its own model code

1. Install the Hugging Face hub package.
2. Download the model, or load it by name in Transformers.
3. Allow remote code when you load it, as the repo carries custom model code.

pip install -U huggingface_hub
hf download klein-ml/Klein-TTS-X1.5-Pro

How to use it: Load the model with AutoModel, call tts with your text and a voice, then save the waveform to a WAV file. The voices call lists the recommended voices for a language.

from transformers import AutoModel

repo = "klein-ml/Klein-TTS-X1.5-Pro"
model = AutoModel.from_pretrained(repo, trust_remote_code=True).eval()

out = model.tts("Hello from Klein TTS. This is a small model that speaks four languages.", voice="Sarah")
model.save_wav("hello.wav", out.waveform, out.sample_rate)

Tool | What it does
model.tts | Makes speech from your text with a chosen voice
model.save_wav | Saves the waveform as a WAV file
model.voices | Lists the recommended voices for a language

Example: You ask for a short English sentence in the voice Sarah. The model returns a waveform, and you save it as a 24 kHz WAV file.

For newcomers and experts: If you are new, run the short script above and listen to the file. Then try another voice or language.

The generator turns phonemes into audio. A text encoder first predicts how long each sound lasts, then the pitch and loudness. A decoder built around a harmonic sound source and an inverse STFT output stage then produces the audio. Training used distillation from synthetic speech, with a GAN objective and a feature loss from WavLM.

Text becomes phonemes
The encoder predicts durations
Pitch and energy are predicted
The decoder writes the waveform
You save a 24 kHz WAV file

Good to know:

• The card reports 4.44 UTMOS and a 2.51 percent word error rate on 300 held-out sentences
• The 82M Kokoro model scores higher on UTMOS and SCOREQ
• Loading uses trust_remote_code, so review the repo code first
• License: Apache 2.0
• Cost to run: none named
• Data it reaches: the text you send and the files downloaded from Hugging Face
• Not for: the card names no banned uses

Title: Parameters across the X1.5 family
Flash Lite: 0.4 million
Flash: 1.7 million
Pro: 13 million

On Hugging Face, 13 November 2026

• Task: text to speech
• Library: pytorch
• Parameters: 13.1M
• Weights: about 52 MB in F32; about 7 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 258
• Likes: 10
• License: apache-2.0
• Languages: en, es, fr, it
• Tags: pytorch, onnx, safetensors, klein_tts, text-to-speech, multilingual, distillation, on-device, kleinml, klein-tts

Links

• Model: https://huggingface.co/klein-ml/Klein-TTS-X1.5-Pro
• Files: https://huggingface.co/klein-ml/Klein-TTS-X1.5-Pro/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download klein-ml/Klein-TTS-X1.5-Pro

Then follow the usage in its model card. Nothing in this kit runs by itself.