snakers4/silero-models

• Silero Models
  • Installation and Basics
  • Text-To-Speech
    • Models and Speakers
      • V5 Turkic
      • V5 Caucasian
      • V5
      • V5 CIS Base Models
      • V5 CIS Ext Models
      • V4
      • V3
    • Dependencies
    • PyTorch
    • Standalone Use
    • SSML
    • Cyrillic languages v4
    • Indic languages v4
      • Example
      • Supported languages
  • Contact
  • Licence
  • Citations
  • Further reading
    • English
    • Chinese
    • Russian

Silero Models

Our TTS models satisfy the following criteria:

• Fully end-to-end;
• Large library of voices;
• Natural-sounding speech;
• One-line usage, minimal, portable;
• Impressively fast on CPU and GPU;
• For the Russian language - automated stress and homographs;

Installation and Basics

You can basically use our models in 3 flavours:

• Via PyTorch Hub: torch.hub.load();
• Via pip: pip install silero and then from silero import silero_tts;
• Via caching the required models and utils manually and modifying if necessary;

Models are downloaded on demand both by pip and PyTorch Hub. If you need caching, do it manually or via invoking a necessary model once (it will be downloaded to a cache folder). Please see these docs for more information.

PyTorch Hub and pip package are based on the same code. All of the torch.hub.load examples can be used with the pip package via this basic change:

from silero import silero_tts
model, example_text = silero_tts(language='ru',
                                 speaker='v5_ru')
audio = model.apply_tts(text=example_text)

Text-To-Speech

Models and Speakers

All of the provided models are listed in the models.yml file. Any metadata and newer versions will be added there.

V5 Turkic

• This model does not support stress at all;
• This model supports SSML;
• This model supports 8000, 24000, 48000 sampling rates;

ID	Speakers	Language	Colab
v5_turkic	chv_0,chv_1	chv (Chuvash)
v5_turkic	crh_0, crh_1, crh_2	crh (Crimean Tatar)
v5_turkic	gag_0, gag_1, gag_2	gag (Gagauz)
v5_turkic	kaa_0, kaa_1	kaa (Karakalpak)
v5_turkic	kir_0	kir (Kyrgyz)
v5_turkic	kjh_0	kjh (Khakas)
v5_turkic	sah_0, sah_1, sah_2	sah (Yakut )
v5_turkic	sty_0	sty (Siberian Tatar)
v5_turkic	tat_0	tat (Tatar)
v5_turkic	tgk_0	tgk (Tajik)
v5_turkic	tuk_0, tuk_1, tuk_2	tuk (Turkmen)
v5_turkic	tyv_0	tyv (Tuvan)
v5_turkic	uzb_0	uzb (Uzbek)
v5_turkic	xal_0, xal_1, xal_2	xal (Kalmyk)

V5 Caucasian

• This model does not support stress at all;
• This model supports SSML;
• This model supports 8000, 24000, 48000 sampling rates;

ID	Speakers	Language	Colab
v5_caucasian	abq_0, abq_1, abq_2	abq (Abaza)
v5_caucasian	ady_0, ady_1, ady_2, ady_3	ady (Adygean)
v5_caucasian	agx_0	agx (Aghul)
v5_caucasian	ava_0	ava (Avar)
v5_caucasian	che_0, che_1, che_2, che_3, che_4	che (Chechen)
v5_caucasian	darg_0, darg_1	darg (Dargin)
v5_caucasian	inh_0, inh_1	inh (Ingush)
v5_caucasian	kbd_0, kbd_1, kbd_2	kbd (K.-Ciscassian)
v5_caucasian	krc_0, krc_1, krc_2	krc (K.-Balkar)
v5_caucasian	kum_0, kum_1, kum_2	kum (Kumyk)
v5_caucasian	lbe_0	lbe (Lak)
v5_caucasian	lez_0	lez (Lezgin)
v5_caucasian	oss_0	oss (Ossetian)
v5_caucasian	tab_0, tab_1	tab (Tabasaran)
v5_caucasian	tkr_0	tkr (Tsakhur)

examples_tts_turkic_caucasian.ipynb

V5

V5 models support SSML. Also see Colab examples for main SSML tag usage.

Russian-only models support automated stress and homographs. v5_2_ru cointains minor fixes and removes numpy and scipy dependencies.

v5_3_ru cointains minor fixes. v5_4_ru also supports questions.

ID	Speakers	Auto-stress / Homographs / Questions	Language	SR	Colab
v5_5_ru	aidar, baya, kseniya, xenia, eugene	✅ / ✅ / ✅	ru (Russian)	8000, 24000, 48000
v5_4_ru	aidar, baya, kseniya, xenia	✅ / ✅ / ✅	ru (Russian)	8000, 24000, 48000
v5_3_ru	aidar, baya, kseniya, xenia, eugene	✅ / ✅ / ❌	ru (Russian)	8000, 24000, 48000
v5_2_ru	aidar, baya, kseniya, xenia, eugene	✅ / ✅ / ❌	ru (Russian)	8000, 24000, 48000
v5_ru	aidar, baya, kseniya, xenia, eugene	✅ / ✅ / ❌	ru (Russian)	8000, 24000, 48000

V5 CIS Base Models

• All of the below models support 8000, 24000, 48000 sampling rates and contain no auto-stress or homographs;
• v5_cis_base models assume that proper stress should be added for each word for all languages, i.e. к+ошка;
• v5_cis_base_nostress models assume that proper stress should be added for each word ONLY for slavic languages (i.e. ru, bel, ukr);
• All of the below models are published under MIT licence;
• V5 UTMOS and throughput metrics;
• V5 models support SSML. Also see Colab examples for main SSML tag usage;
• Use cases for the model;
• Minimal system requirements: a PyTorch-compatible system, a modern processor with AVX2 instruction set for x86/64 platform.

ID	Speakers	Language	Colab
v5_cis_base, v5_cis_base_nostress	aze_gamat	aze (Azerbaijani)
v5_cis_base, v5_cis_base_nostress	hye_zara	hye (Armenian)
v5_cis_base, v5_cis_base_nostress	bak_aigul, bak_alfia, bak_alfia2	bak (Bashkir)
v5_cis_base, v5_cis_base_nostress	bak_miyau, bak_ramilia	bak (Bashkir)
v5_cis_base, v5_cis_base_nostress	bel_anatoliy, bel_dmitriy, bel_larisa	bel (Belarus)
v5_cis_base, v5_cis_base_nostress	kat_vika	kat (Georgian)
v5_cis_base, v5_cis_base_nostress	kbd_eduard	kbd (Kab.-Cherkes)
v5_cis_base, v5_cis_base_nostress	kaz_zhadyra, kaz_zhazira	kaz (Kazakh)
v5_cis_base, v5_cis_base_nostress	xal_kejilgan, xal_kermen	xal (Kalmyk)
v5_cis_base, v5_cis_base_nostress	kir_nurgul	kir (Kyrgyz)
v5_cis_base, v5_cis_base_nostress	mdf_oksana	mdf (Moksha)
v5_cis_base, v5_cis_base_nostress	all of these speakers, but with ru_ prefix	ru (Russian)
v5_cis_base, v5_cis_base_nostress	tgk_onaoy, tgk_safarhuja	tgk (Tajik)
v5_cis_base, v5_cis_base_nostress	tat_albina, tat_marat	tat (Tatar)
v5_cis_base, v5_cis_base_nostress	udm_bogdan	udm (Udmurt)
v5_cis_base, v5_cis_base_nostress	uzb_saida	uzb (Uzbek)
v5_cis_base, v5_cis_base_nostress	ukr_igor, ukr_roman	ukr (Ukrainian)
v5_cis_base, v5_cis_base_nostress	kjh_karina, kjh_sibday	kjh (Khakas)
v5_cis_base, v5_cis_base_nostress	chv_ekaterina	chv (Chuvash)
v5_cis_base, v5_cis_base_nostress	erz_alexandr	erz (Erzya)
v5_cis_base, v5_cis_base_nostress	sah_zinaida	sah (Yakut)

V5 CIS Ext Models

• All of the below models support 8000, 24000, 48000 sampling rates and contain no auto-stress or homographs;
• v5_cis_ext models assume that proper stress should be added for each word for all languages, i.e. к+ошка;
• v5_cis_ext_nostress are coming soon;
• All of the below models are published under CC-NC-BY licence;
• V5 models support SSML. Also see Colab examples for main SSML tag usage.

ID	Speakers	Language	Colab
v5_cis_ext	kaz_abai, kaz_aidana, kaz_aisha, kaz_bakir, kaz_danara	kaz (Kazakh)
v5_cis_ext	xal_delghir, xal_erdni	xal (Kalmyk)
v5_cis_ext	tat_adiba, tat_alsou, tat_amir, tat_azat, tat_batir	tat (Tatar)
v5_cis_ext	tat_bulat, tat_damir, tat_guzel, tat_ildar, tat_ilgiz	tat (Tatar)
v5_cis_ext	tat_karim, tat_mansur, tat_murat, tat_rasima, tat_rustem	tat (Tatar)
v5_cis_ext	tat_timur, tat_zifa, tat_zufar, tat_zulfiya	tat (Tatar)
v5_cis_ext	uzb_anora, uzb_dilnavoz	uzb (Uzbek)
v5_cis_ext	ukr_kateryna, ukr_lada, ukr_mykyta, ukr_oleksa, ukr_tetiana	ukr (Ukrainian)
v5_cis_ext	chv_aihwa, chv_alima	chv (Chuvash)

V4

V4 models support SSML. Also see Colab examples for main SSML tag usage.

V4 models: v4_ru, v4_cyrillic, v4_ua, v4_uz, v4_indic

ID	Speakers	Auto-stress	Language	SR	Colab
v4_ru	aidar, baya, kseniya, xenia, eugene, random	yes	ru (Russian)	8000, 24000, 48000
v4_cyrillic	b_ava, marat_tt, kalmyk_erdni...	no	cyrillic (Avar, Tatar, Kalmyk, ...)	8000, 24000, 48000
v4_ua	mykyta, random	no	ua (Ukrainian)	8000, 24000, 48000
v4_uz	dilnavoz	no	uz (Uzbek)	8000, 24000, 48000
v4_indic	hindi_male, hindi_female, ..., random	no	indic (Hindi, Telugu, ...)	8000, 24000, 48000

V3

V3 models support SSML. Also see Colab examples for main SSML tag usage.

V3 models: v3_en, v3_en_indic, v3_de, v3_es, v3_fr, v3_indic

ID	Speakers	Auto-stress	Language	SR	Colab
v3_en	en_0, en_1, ..., en_117, random	no	en (English)	8000, 24000, 48000
v3_en_indic	tamil_female, ..., assamese_male, random	no	en (English)	8000, 24000, 48000
v3_de	eva_k, ..., karlsson, random	no	de (German)	8000, 24000, 48000
v3_es	es_0, es_1, es_2, random	no	es (Spanish)	8000, 24000, 48000
v3_fr	fr_0, ..., fr_5, random	no	fr (French)	8000, 24000, 48000
v3_indic	hindi_male, hindi_female, ..., random	no	indic (Hindi, Telugu, ...)	8000, 24000, 48000

Dependencies

Basic dependencies for Colab examples:

• torch, 1.10+ for v3 models/ 2.0+ for v4 and v5 models;
• torchaudio, latest version bound to PyTorch should work (required only because models are hosted together with STT, not required for work);
• omegaconf, latest (can be removed as well, if you do not load all of the configs);

PyTorch

# V5
import torch

language = 'ru'
model_id = 'v5_ru'
sample_rate = 48000
speaker = 'xenia'
device = torch.device('cpu')

model, example_text = torch.hub.load(repo_or_dir='snakers4/silero-models',
                                     model='silero_tts',
                                     language=language,
                                     speaker=model_id)
model.to(device)  # gpu or cpu

audio = model.apply_tts(text=example_text,
                        speaker=speaker,
                        sample_rate=sample_rate)

Standalone Use

• Standalone usage only requires PyTorch 1.12+ and the Python Standard Library;
• Please see the detailed examples in Colab;

# V5
import os
import torch

device = torch.device('cpu')
torch.set_num_threads(4)
local_file = 'model.pt'

if not os.path.isfile(local_file):
    torch.hub.download_url_to_file('https://models.silero.ai/models/tts/ru/v5_ru.pt',
                                   local_file)  

model = torch.package.PackageImporter(local_file).load_pickle("tts_models", "model")
model.to(device)

example_text = 'Меня зовут Лева Королев. Я из готов. И я уже готов открыть все ваши замки любой сложности!'
sample_rate = 48000
speaker='baya'

audio_paths = model.save_wav(text=example_text,
                             speaker=speaker,
                             sample_rate=sample_rate)

SSML

Check out our TTS Wiki page.

Cyrillic languages v4

To be superseded with v5 model(s) soon.

Supported tokenset: !,-.:?iµöабвгдежзийклмнопрстуфхцчшщъыьэюяёђѓєіјњћќўѳғҕҗҙқҡңҥҫүұҳҷһӏӑӓӕӗәӝӟӥӧөӱӳӵӹ

Indic languages v4

Example

(!!!) All input sentences should be romanized to ISO format using aksharamukha. An example for hindi:

# V3
import torch
from aksharamukha import transliterate

# Loading model
model, example_text = torch.hub.load(repo_or_dir='snakers4/silero-models',
                                     model='silero_tts',
                                     language='indic',
                                     speaker='v4_indic')

orig_text = "प्रसिद्द कबीर अध्येता, पुरुषोत्तम अग्रवाल का यह शोध आलेख, उस रामानंद की खोज करता है"
roman_text = transliterate.process('Devanagari', 'ISO', orig_text)
print(roman_text)

audio = model.apply_tts(roman_text,
                        speaker='hindi_male')

Supported languages

Contact

Try our models, create an issue, join our chat, email us, and read the latest news.

Licence

All of the models are published under the main repo license (i.e. CC-NC-BY) except for the base cis-tts models, which are under MIT.

Citations

@misc{Silero Models,
  author = {Silero Team},
  title = {Silero Models: pre-trained text-to-speech models made embarrassingly simple},
  year = {2025},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/snakers4/silero-models}},
  commit = {insert_some_commit_here},
  email = {[email protected]}
}

Further reading

English

• STT:

  • Towards an Imagenet Moment For Speech-To-Text - link
  • A Speech-To-Text Practitioners Criticisms of Industry and Academia - link
  • Modern Google-level STT Models Released - link

• TTS:

  • Multilingual Text-to-Speech Models for Indic Languages - link
  • Our new public speech synthesis in super-high quality, 10x faster and more stable - link
  • High-Quality Text-to-Speech Made Accessible, Simple and Fast - link

• VAD:

  • One Voice Detector to Rule Them All - link
  • Modern Portable Voice Activity Detector Released - link

• Text Enhancement:

  • We have published a model for text repunctuation and recapitalization for four languages - link

Chinese

• STT:
  • 迈向语音识别领域的 ImageNet 时刻 - link
  • 语音领域学术界和工业界的七宗罪 - link

Russian

• STT

  • OpenAI решили распознавание речи! Разбираемся так ли это … - link
  • Наши сервисы для бесплатного распознавания речи стали лучше и удобнее - link
  • Telegram-бот Silero бесплатно переводит речь в текст - link
  • Бесплатное распознавание речи для всех желающих - link
  • Последние обновления моделей распознавания речи из Silero Models - link
  • Сжимаем трансформеры: простые, универсальные и прикладные способы cделать их компактными и быстрыми - link
  • Ультимативное сравнение систем распознавания речи: Ashmanov, Google, Sber, Silero, Tinkoff, Yandex - link
  • Мы опубликовали современные STT модели сравнимые по качеству с Google - link
  • Понижаем барьеры на вход в распознавание речи - link
  • Огромный открытый датасет русской речи версия 1.0 - link
  • Насколько Быстрой Можно Сделать Систему STT? - link
  • Наша система Speech-To-Text - link
  • Speech-To-Text - link

• TTS:

  • Теперь silero-tts v5 на русском языке умеет задавать вопросы - link
  • Наш синтез для 20 языков теперь работает локально под Windows как экранная читалка (SAPI5) и в Балаболке - link
  • Мы добавили поддержку ещё 19 языков России и СНГ в проект silero-stress - link
  • Мы опубликовали стабильный, быстрый, качественный и доступный синтез для 20 языков России - link
  • Мы опубликовали silero-tts v5 на русском языке - link
  • Мы решили задачу омографов и ударений в русском языке - link
  • Делаем быстрый, качественный и доступный синтез на языках России — нужно ваше участие - link
  • Теперь наш синтез также доступен в виде бота в Телеграме - link
  • Может ли синтез речи обмануть систему биометрической идентификации? - link
  • Теперь наш синтез на 20 языках - link
  • Теперь наш публичный синтез в супер-высоком качестве, в 10 раз быстрее и без детских болячек - link
  • Синтезируем голос бабушки, дедушки и Ленина + новости нашего публичного синтеза - link
  • Мы сделали наш публичный синтез речи еще лучше - link
  • Мы Опубликовали Качественный, Простой, Доступный и Быстрый Синтез Речи - link

• VAD:

  • Новый релиз публичного детектора голоса Silero VAD v6 - link
  • Наш публичный детектор голоса стал лучше - link
  • А ты используешь VAD? Что это такое и зачем он нужен - link
  • Модели для Детекции Речи, Чисел и Распознавания Языков - link
  • Мы опубликовали современный Voice Activity Detector и не только -link

• Text Enhancement:

  • Восстановление знаков пунктуации и заглавных букв — теперь и на длинных текстах - link
  • Мы опубликовали модель, расставляющую знаки препинания и заглавные буквы в тексте на четырех языках - link