You can deploy this model and start earning money today!

MuzaffarSharofitdinov/
mms-tts-uzbek-finetuned

Name: mms-tts-uzbek-finetuned
Availability: InStock

safetensors

vits

Views

No views yet

MMS TTS Uzbek — Fine-tuned

facebook/mms-tts-uzb-script_cyrillic modelini o'zbek datasetida fine-tune qilingan versiyasi.

Ishlatish

python

import torch
import soundfile as sf
import numpy as np
import scipy.signal as signal
from transformers import VitsModel, AutoTokenizer
from IPython.display import Audio, display

# Model yuklash
model = VitsModel.from_pretrained("MuzaffarSharofitdinov/mms-tts-uzbek-finetuned")
tokenizer = AutoTokenizer.from_pretrained("MuzaffarSharofitdinov/mms-tts-uzbek-finetuned")
model.eval()

# Lotin → Kiril konvertatsiya
LAT2CYR = {
    "sh": "ш", "ch": "ч", "ng": "нг",
    "g'": "ғ", "o'": "ў",
    "yo": "ё", "yu": "ю", "ya": "я", "ts": "ц",
    "a": "а", "b": "б", "v": "в", "g": "г", "d": "д",
    "e": "е", "z": "з", "i": "и", "y": "й", "k": "к",
    "l": "л", "m": "м", "n": "н", "o": "о", "p": "п",
    "r": "р", "s": "с", "t": "т", "u": "у", "f": "ф",
    "x": "х", "q": "қ", "h": "ҳ", "j": "ж", " ": " ",
}

def latin_to_cyrillic(text):
    if not isinstance(text, str):
        return ''
    text = text.lower()
    result = ''
    i = 0
    while i < len(text):
        if i + 1 < len(text) and text[i:i+2] in LAT2CYR:
            result += LAT2CYR[text[i:i+2]]
            i += 2
        elif text[i] in LAT2CYR:
            result += LAT2CYR[text[i]]
            i += 1
        else:
            i += 1
    return result

# Shovqin filtri
def remove_noise(audio, sample_rate=16000):
    b, a = signal.butter(2, 7000 / (sample_rate / 2), btype='low')
    audio_filtered = signal.filtfilt(b, a, audio)
    audio_filtered = audio_filtered / (np.max(np.abs(audio_filtered)) + 1e-8) * 0.95
    return audio_filtered.astype(np.float32)

# TTS funksiyasi
def speak(text_latin, filename='output.wav'):
    text_cyr = latin_to_cyrillic(text_latin)
    inputs = tokenizer(text_cyr, return_tensors='pt')
    with torch.no_grad():
        audio = model(
            **inputs,
            noise_scale=0.1,
            noise_scale_w=0.5,
            length_scale=1.2,
        ).waveform.squeeze().detach().numpy()
    audio = remove_noise(audio, model.config.sampling_rate)
    sf.write(filename, audio, model.config.sampling_rate)
    return audio

# Test
speak("salom, bu sinov ovozi", "output.wav")
display(Audio('output.wav'))

Model haqida

Asos model: facebook/mms-tts-uzb-script_cyrillic
Dataset: Beehzod/uzbek_stt_data (Common Voice o'zbek)
Yozuv: Kirilcha (lotin matn avtomatik konvertatsiya qilinadi)
Training: Freeze usuli — faqat flow va decoder qatlamlari

Parametrlar

Parametr	Qiymat	Tavsif
`noise_scale`	0.1	Past = tekisroq ovoz
`noise_scale_w`	0.5	Ritm barqarorligi
`length_scale`	1.2	Ovoz tezligi (1.0 = normal)