---
language: tr
license: cc-by-4.0
tags:
- text-to-speech
- tts
- piper
- vits
- audio
- turkish
metrics:
- wer
pipeline_tag: text-to-speech
widget:
- text: "Merhaba, bu Piper TTS Türkçe xlow ses modelinin bir testidir."
  example_title: Example 1
- text: "Yapay zeka ve derin öğrenme teknolojileri her geçen gün daha da gelişiyor."
  example_title: Example 2
---

# Piper TTS: Turkish (tr_TR) - xlow Quality

This model is a highly lightweight and optimized **xlow** (ultra-low resource / high performance) version trained on the **Piper TTS** architecture for the Turkish (`tr_TR`) language. It is specifically designed for embedded systems, low-spec hardware, or real-time applications requiring ultra-low latency inference.

## Model Details

* **Developer:** [dcx514ai](https://huggingface.co/dcx514ai)
* **Language:** Turkish (tr_TR)
* **Model Type:** VITS / Piper TTS
* **Quality Level:** xlow (Ultra Low)
* **Training Checkpoint:** `epoch=904-step=1826290`

**Technical Specifications:** 24 kHz Sample Rate | 256 Phoneme Embeddings | Single-Speaker | 1.7 Million Inference Parameters (4.3 Million Total Training Parameters)

## Repository Structure & Files

The repository contains the following artifacts to support deployment across multiple platforms and runtime environments:

| File Name | Size | Description |
| :--- | :--- | :--- |
| `ep0904_st1826290_20260710_225050_last.onnx` | 7.36 MB | Standard ONNX Export (Piper compatible) |
| `ep0904_st1826290_20260710_225050_last.q8.onnx` | 3.43 MB | **8-bit Quantized** ONNX Export (Maximum speed and footprint optimization) |
| `ep0904_st1826290_20260710_225050_last.onnx.json` | 5.01 kB | Model configuration and phoneme dictionary file |
| `ep0904_st1826290_20260710_225050_last.pt` | 11.20 MB | PyTorch Model Weights |
| `epoch=904-step=1826290.ckpt` | 55.10 MB | PyTorch Lightning Training Checkpoint |

---

## Audio Sample

You can preview the synthesized audio quality directly via the player below:
# Sample
<audio controls src="https://huggingface.co/dcx514ai/piper_tts_turkish_xlow/resolve/main/sample_audio.wav"></audio>

# Transcript
Genel ekonomide fiyatlar, beklentiler ve güven duygusu birlikte hareket eder. Örneğin, bir bankanın CEO'su faiz kararını açıklarken yalnızca rakamları 
değil piyasadaki algıyı, şirketlerin yatırım iştahını ve hane halkının harcama eğilimini de dikkate alır. Kısa vadede dalgalanmalar görülebilir, ancak orta ve 
uzun vadede istikrar; doğru iletişim, öngörülebilir politika ve sağlam bilanço yönetimiyle güçlenir.

---

## Usage Guide

### 1. Direct Execution via Piper CLI

Once you download the matching `.onnx` and `.onnx.json` files, you can run synthesis from your terminal as follows:

```bash
echo "Piper Türkçe xlow modeli başarıyla çalışıyor." | \
  piper --model ep0904_st1826290_20260710_225050_last.onnx \
        --config ep0904_st1826290_20260710_225050_last.onnx.json \
        --output_file output.wav