---
license: apache-2.0
base_model: HKUSTAudio/Llasa-1B-Multilingual
datasets:
  - amu-cai/pl-asr-bigos-v2
language: 
  - pl
tags:
  - speech
  - audio
  - polish
  - llama
  - tts
  - fine-tuned
  - text-to-speech
model-index:
  - name: From Llasa to Łazanki
    results: []
---

# From Llasa to Łazanki: Fine-tuned Llasa-1B on Polish Speech

This is a fine-tuned version of [`HKUSTAudio/Llasa-1B-Multilingual`](https://huggingface.co/HKUSTAudio/Llasa-1B-Multilingual), adapted for **Polish Text-to-Speech (TTS)**.  
It was fine-tuned on the [`pl-asr-bigos-v2`](https://huggingface.co/datasets/amu-cai/pl-asr-bigos-v2) dataset, specifically the `mozilla-common_voice_15-23` subset, which includes high-quality Polish speech recordings suitable for training TTS models.

---

## 🧠 Base Model

[Llasa-1B-Multilingual](https://huggingface.co/HKUSTAudio/Llasa-1B-Multilingual) model developed by HKUST. The approach leverages the LLAMA-initialized text BPE tokenizer, which can handle multilingual text without the need to design language-specific G2P (grapheme-to-phoneme) systems.

---

## 🗣 Fine-tuning Details

- **Dataset**: [PL-ASR-BIGOS-v2](https://huggingface.co/datasets/amu-cai/pl-asr-bigos-v2), `mozilla-common_voice_15-23` subset  
- **Language**: 🇵🇱 Polish
- **Task**: Text to speech