Qwen3 TTS Tiny Turkish

Qwen3 TTS Tiny Turkish

Speaks Turkish text aloud in a voice chosen from a short reference recording. A 199M text to speech model built on Qwen3-TTS, small enough to run on a laptop.

Made by erkamk 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: This model reads Turkish text aloud. It is compact, with 199M parameters. It keeps the structure of Qwen3-TTS-12Hz-0.6B-Base but is about 4.6 times smaller. You pick the voice by giving it a clean reference recording of 5 to 10 seconds in Turkish. It needs no per voice fine tuning and no fixed list of voices.

Who it suits: Developers, creators and researchers who need Turkish speech from text on modest hardware, and who have the right to use the voices they supply.

Industry | Who uses it | How they use it | Example
Media and content | Creators who voice Turkish videos | Generate narration from a script in a chosen voice | Reading a short article aloud
Software development | Developers who add speech to apps | Run a small model locally for Turkish voice output | Adding spoken replies to a Turkish assistant
Research | People who study speech synthesis | Test a compact model against larger ones | Comparing speaker similarity with a larger Turkish model

Skip it if you need a language other than Turkish, or streaming output, since the card lists neither. Only use voices you have the right to use.

Where to use it: It uses the qwen-tts library, on a GPU or a laptop CPU. The card says a MacBook CPU runs it faster than real time. The files download on Windows, macOS or Linux.

• Python with the qwen-tts library
• A GPU, or a CPU with under 4 GB of RAM free on a MacBook according to the card
• Windows, macOS or Linux: where you can fetch the files with Python

Use case | Who | What they get back
Narration | Creators | Spoken Turkish audio at 24 kHz
Voice choice from a sample | Developers | Speech in the voice of a short reference recording
Small model research | Researchers | A compact model to inspect and compare

How to set it up: You need Python and the qwen-tts library.

• Python with pip
• The qwen-tts library
• About 399 MB of space for the BF16 weights
• A clean Turkish reference recording of 5 to 10 seconds, which you have the right to use

1. Install the Hugging Face hub library.
2. Download the model files.
3. Load them with qwen-tts and give a reference recording.

pip install -U huggingface_hub
hf download erkamk/qwen3-tts-tiny-tr

How to use it: The part of the card shown here gives no run command. Supply Turkish text and a reference recording, and the model speaks the text in that voice. Output is not streamed.

Tool | What it does
hf download | Fetches the model files to your computer
qwen-tts | Loads the model and produces speech from text

Example: You give it a Turkish sentence about a ferry crossing in Istanbul and a 7 second recording of a voice you may use. It returns an audio file of the sentence spoken in that voice.

For newcomers and experts: If you are new to this, start with one short sentence and a clean recording with no background noise. Listen to how closely the voice matches.

If you know the field, only the talker was trained, with 8 layers of width 512 against the base's 28 of 1024. Training was supervised, then reinforcement learning on speaker similarity and naturalness. The speech tokenizer is unchanged.

Turkish text and a reference clip
Speaker encoder reads the voice
Small talker predicts speech codes
Speech tokenizer outputs audio

Good to know: The card reports its own tests on 120 Turkish sentences read in real speakers' voices. The larger Turkish model sounds more natural, while this one follows a new voice more closely.

Title: Card test with real speaker voices
Model | Word error rate | Speaker similarity
This model | 3.3% | 0.46
Qwen3 0.6B Turkish | 2.0% | 0.39

• License: Apache 2.0
• Cost to run: none named
• Data it reaches: the text and reference recordings you provide
• Not for: languages other than Turkish, and voices you have no right to use

On Hugging Face, 13 November 2026

• Task: text to speech
• Library: qwen-tts
• Parameters: 199M
• Weights: about 399 MB in BF16; about 100 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 682
• Likes: 15
• License: apache-2.0
• Languages: tr
• Based on: Qwen/Qwen3-TTS-12Hz-0.6B-Base
• Tags: qwen-tts, safetensors, qwen3_tts, qwen3-tts, turkish, voice-cloning, text-to-speech

Links

• Model: https://huggingface.co/erkamk/qwen3-tts-tiny-tr
• Files: https://huggingface.co/erkamk/qwen3-tts-tiny-tr/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download erkamk/qwen3-tts-tiny-tr

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