Skip to main content
Omnilingual ASR supports over 1,600 languages across multiple scripts, including hundreds of languages never before covered by any ASR system.

​
Language ID Format

Languages in Omnilingual ASR follow a standardized format: 
{language_code}_{script}

​
Format Components

​
language_code
string
required
Three-letter ISO 639-3 language code (e.g., eng for English, spa for Spanish, cmn for Mandarin)
​
script
string
required
Four-letter ISO 15924 script code (e.g., Latn for Latin, Arab for Arabic, Hans for Simplified Chinese)

​
Examples

"eng_Latn"  # English (Latin script)
"spa_Latn"  # Spanish (Latin script)
"fra_Latn"  # French (Latin script)
"deu_Latn"  # German (Latin script)
"cmn_Hans"  # Mandarin Chinese (Simplified)
"cmn_Hant"  # Mandarin Chinese (Traditional)
"jpn_Jpan"  # Japanese
"kor_Hang"  # Korean (Hangul)
"arb_Arab"  # Standard Arabic
"rus_Cyrl"  # Russian (Cyrillic)
"hin_Deva"  # Hindi (Devanagari)

​
Accessing the Language List

You can programmatically access the full list of supported languages: 
from omnilingual_asr.models.wav2vec2_llama.lang_ids import supported_langs

# Print total count
print(f"Total supported languages: {len(supported_langs)}")

# Print all languages
for lang in supported_langs:
    print(lang)

​
Script Coverage

Omnilingual ASR supports multiple writing systems:
Latin Script (Latn) - Most widely usedThe majority of supported languages use Latin script, including:
  • European languages (English, Spanish, French, German, etc.)
  • African languages (Swahili, Yoruba, Hausa, etc.)
  • Southeast Asian languages (Indonesian, Vietnamese, Filipino, etc.)
  • Indigenous American languages (Quechua, Guarani, Nahuatl, etc.)
"eng_Latn", "spa_Latn", "fra_Latn", "deu_Latn", "ita_Latn",
"por_Latn", "pol_Latn", "nld_Latn", "ind_Latn", "vie_Latn"

​
Language Statistics

Total Languages

1,682 language-script combinations

High Performance

78% with CER below 10%

New Coverage

Hundreds of previously uncovered languages

​
Using Language IDs in Code

Here’s how to use language IDs with the inference pipeline: 
from omnilingual_asr.models.inference.pipeline import ASRInferencePipeline

pipeline = ASRInferencePipeline(model_card="omniASR_LLM_Unlimited_7B_v2")

# Transcribe English audio
transcriptions = pipeline.transcribe(
    ["/path/to/english_audio.wav"],
    lang=["eng_Latn"],
    batch_size=1
)

​
Performance by Language

For detailed per-language performance metrics, see the complete results in the per_language_results_table_7B_llm_asr.csv file.
The 7B LLM-ASR model achieves character error rates (CER) below 10% for 78% of supported languages.

​
Language Families

Omnilingual ASR covers languages from diverse language families:
  • Indo-European: Romance, Germanic, Slavic, Indo-Aryan, Iranian languages
  • Sino-Tibetan: Chinese varieties, Tibetan, Burmese
  • Niger-Congo: Bantu, Atlantic, Gur, Kwa languages
  • Austronesian: Indonesian, Filipino, Polynesian languages
  • Afro-Asiatic: Arabic, Amharic, Hausa, Hebrew
  • Dravidian: Tamil, Telugu, Kannada, Malayalam
  • Turkic: Turkish, Uzbek, Kazakh, Uyghur
  • Uralic: Finnish, Hungarian, Estonian
  • Indigenous American: Quechua, Guarani, Nahuatl, Aymara
  • And many more…

​
Finding Your Language

To find the correct language ID:
1

Identify ISO 639-3 Code

Look up your language’s three-letter code at ISO 639-3 or Ethnologue.Example: English = eng, Spanish = spa
2

Determine Script

Identify which script your text uses:
  • Latin alphabet → Latn
  • Cyrillic → Cyrl
  • Arabic → Arab
  • See ISO 15924 for complete list
3

Combine and Validate

Combine as {language}_{script} and check if it’s in the supported list:
from omnilingual_asr.models.wav2vec2_llama.lang_ids import supported_langs

lang_id = "eng_Latn"
if lang_id in supported_langs:
    print(f"✓ {lang_id} is supported")

​
Complete Language List

The complete list of 1,682 supported languages is available in the source code at: 
src/omnilingual_asr/models/wav2vec2_llama/lang_ids.py
A few example languages from the complete list: 
# Sample of supported languages
"aae_Latn",  # Arbëreshë Albanian
"aka_Latn",  # Akan
"amh_Ethi",  # Amharic
"arb_Arab",  # Standard Arabic
"asm_Beng",  # Assamese
"bam_Latn",  # Bambara
"ben_Beng",  # Bengali
"bul_Cyrl",  # Bulgarian
"cat_Latn",  # Catalan
"ces_Latn",  # Czech
"cmn_Hans",  # Mandarin Chinese (Simplified)
"cmn_Hant",  # Mandarin Chinese (Traditional)
"dan_Latn",  # Danish
"deu_Latn",  # German
"ell_Grek",  # Greek
"eng_Latn",  # English
"epo_Latn",  # Esperanto
"eus_Latn",  # Basque
"fas_Arab",  # Persian
"fin_Latn",  # Finnish
"fra_Latn",  # French
"gle_Latn",  # Irish
"glg_Latn",  # Galician
"grn_Latn",  # Guarani
"guj_Gujr",  # Gujarati
"hat_Latn",  # Haitian Creole
"hau_Latn",  # Hausa
"heb_Hebr",  # Hebrew
"hin_Deva",  # Hindi
"hrv_Latn",  # Croatian
"hun_Latn",  # Hungarian
"hye_Armn",  # Armenian
"ibo_Latn",  # Igbo
"ind_Latn",  # Indonesian
"isl_Latn",  # Icelandic
"ita_Latn",  # Italian
"jav_Latn",  # Javanese
"jpn_Jpan",  # Japanese
"kan_Knda",  # Kannada
"kat_Geor",  # Georgian
"kaz_Cyrl",  # Kazakh
"khm_Khmr",  # Khmer
"kin_Latn",  # Kinyarwanda
"kor_Hang",  # Korean
"lao_Laoo",  # Lao
"lat_Latn",  # Latin
"lav_Latn",  # Latvian
"lit_Latn",  # Lithuanian
"lug_Latn",  # Luganda
"mal_Mlym",  # Malayalam
"mar_Deva",  # Marathi
"mkd_Cyrl",  # Macedonian
"mlg_Latn",  # Malagasy
"mon_Cyrl",  # Mongolian
"mri_Latn",  # Maori
"mya_Mymr",  # Burmese
"nep_Deva",  # Nepali
"nld_Latn",  # Dutch
"nob_Latn",  # Norwegian Bokmål
"nya_Latn",  # Chichewa
"ori_Orya",  # Odia
"pan_Guru",  # Punjabi
"pol_Latn",  # Polish
"por_Latn",  # Portuguese
"pus_Arab",  # Pashto
"ron_Latn",  # Romanian
"rus_Cyrl",  # Russian
"sin_Sinh",  # Sinhala
"slk_Latn",  # Slovak
"slv_Latn",  # Slovenian
"sna_Latn",  # Shona
"som_Latn",  # Somali
"spa_Latn",  # Spanish
"sqi_Latn",  # Albanian
"srp_Cyrl",  # Serbian
"sun_Latn",  # Sundanese
"swe_Latn",  # Swedish
"swh_Latn",  # Swahili
"tam_Taml",  # Tamil
"tel_Telu",  # Telugu
"tgk_Cyrl",  # Tajik
"tha_Thai",  # Thai
"tir_Ethi",  # Tigrinya
"tuk_Latn",  # Turkmen
"tur_Latn",  # Turkish
"ukr_Cyrl",  # Ukrainian
"urd_Arab",  # Urdu
"uzb_Latn",  # Uzbek
"vie_Latn",  # Vietnamese
"wol_Latn",  # Wolof
"xho_Latn",  # Xhosa
"ydd_Hebr",  # Yiddish
"yor_Latn",  # Yoruba
"zho_Hans",  # Chinese (Simplified)
"zul_Latn",  # Zulu
# ... and 1,600+ more!

​
Next Steps

Quick Start

Start transcribing with language codes

Language Conditioning

Improve accuracy with language hints

Model Selection

Choose the right model for your languages

Zero-Shot Learning

Add new languages with examples

Build docs developers (and LLMs) love

Get started for freeTalk to us