Hebrish TTS-V1

An early, experimental single-speaker fine-tune of Qwen3-TTS-12Hz-1.7B-Base on one voice (Daniel Rosehill, Irish-accented English with Hebrew words mixed in: "Hebrish").

This is a first test checkpoint, not a finished model. It was trained to see what is possible with a small amount of data. Expect rough edges.

Training

Data	Hebrish/Hebrish-TTS-Voice-Recordings, train split: 657 read takes, 50 min after trimming, 24 kHz
Method	Full SFT, upstream finetuning/sft_12hz.py with fixes from QwenLM/Qwen3-TTS PR #278 (label shift, sub-talker), text_projection applied in training, fp32 master weights with bf16 autocast
Schedule	LR 7e-6, 3 epochs, batch 2 x grad. accumulation 4
Hardware	1 x L40S (Modal)

Numbers are spelled out in the training text, so write them out in prompts too ("four hundred and twelve").

Usage

import torch, soundfile as sf
from qwen_tts import Qwen3TTSModel

tts = Qwen3TTSModel.from_pretrained("Hebrish/TTS-V1", device_map="cuda:0", dtype=torch.bfloat16)
wavs, sr = tts.generate_custom_voice(text="Yalla, I'll ring you after the match.", speaker="daniel", language="English")
sf.write("out.wav", wavs[0], sr)

Requires qwen-tts (tested with 0.1.1).

Licence

CC BY-NC 4.0, matching the dataset. The base model is Apache 2.0.