antalia-mini

Antalia-2 Mini is a small Turkish text-to-speech model with one synthetic voice: 7.6M parameters, 48 kHz output, well faster than real time on a CPU; code and weights are Apache-2.0. It takes raw Turkish text; numbers, dates, times, amounts, initialisms, web addresses and English or brand words are normalised inside the package. The voice is not a real person's, and no recordings of real speakers were used in training.

pip install antalia-mini

from antalia_mini import Antalia

tts = Antalia()  # downloads the weights from Hugging Face once; GPU if available
speech = tts.say("Randevunuz 14 Ekim Salı günü saat 15:30'da.", path="randevu.wav", seed=0)
speech.audio, speech.sample_rate, speech.duration, speech.seed

for chunk in tts.stream("Uzun bir metin. Cümle cümle üretilir."):
    ...  # float32 numpy chunks, first sentence first

tts8k = Antalia(sample_rate=8000)  # 48000 (native), 24000, 16000 or 8000

Command line: antalia-mini "Merhaba." -o merhaba.wav

Model card, samples, evaluation and limitations: https://huggingface.co/cloud0day3/antalia-mini

Please tell listeners when they are hearing a synthetic voice, and do not use the model to impersonate real people.