#ASR Models

#DolphinAsr Series

Notes

• License: Apache 2.0
• opt: Optimized version, moves audio feature extraction module out of the model to reduce inference overhead
• Full language and region code mapping:

zh-CN: Chinese (Mandarin), zh-TW: Chinese (Taiwan), zh-WU: Chinese (Wu), zh-SICHUAN: Chinese (Sichuan), zh-SHANXI: Chinese (Shanxi), zh-ANHUI: Chinese (Anhui), zh-TIANJIN: Chinese (Tianjin), zh-NINGXIA: Chinese (Ningxia), zh-SHAANXI: Chinese (Shaanxi), zh-HEBEI: Chinese (Hebei), zh-SHANDONG: Chinese (Shandong), zh-GUANGDONG: Chinese (Guangdong), zh-SHANGHAI: Chinese (Shanghai), zh-HUBEI: Chinese (Hubei), zh-LIAONING: Chinese (Liaoning), zh-GANSU: Chinese (Gansu), zh-FUJIAN: Chinese (Fujian), zh-HUNAN: Chinese (Hunan), zh-HENAN: Chinese (Henan), zh-YUNNAN: Chinese (Yunnan), zh-MINNAN: Chinese (Minnan), zh-WENZHOU: Chinese (Wenzhou)
ja-JP: Japanese, th-TH: Thai, ru-RU: Russian, ko-KR: Korean, id-ID: Indonesian, vi-VN: Vietnamese
ct-NULL: Cantonese, ct-HK: Cantonese (Hong Kong), ct-GZ: Cantonese (Guangdong)
hi-IN: Hindi, ur-IN: Urdu (India), ur-PK: Urdu, ms-MY: Malay, uz-UZ: Uzbek
ar-MA: Arabic (Morocco), ar-GLA: Arabic, ar-SA: Arabic (Saudi Arabia), ar-EG: Arabic (Egypt), ar-KW: Arabic (Kuwait), ar-LY: Arabic (Libya), ar-JO: Arabic (Jordan), ar-AE: Arabic (UAE), ar-LVT: Arabic (Levant)
fa-IR: Persian, bn-BD: Bengali
ta-SG: Tamil (Singapore), ta-LK: Tamil (Sri Lanka), ta-IN: Tamil (India), ta-MY: Tamil (Malaysia)
te-IN: Telugu, ug-NULL: Uyghur, ug-CN: Uyghur, gu-IN: Gujarati
my-MM: Burmese, tl-PH: Tagalog, kk-KZ: Kazakh, or-IN: Odia, ne-NP: Nepali
mn-MN: Mongolian, km-KH: Khmer, jv-ID: Javanese, lo-LA: Lao, si-LK: Sinhala
fil-PH: Filipino, ps-AF: Pashto, pa-IN: Punjabi, kab-NULL: Kabyle
ba-NULL: Bashkir, ks-IN: Kashmiri, tg-TJ: Tajik, su-ID: Sundanese
mr-IN: Marathi, ky-KG: Kyrgyz, az-AZ: Azerbaijani

#DolphinAsr-base Models

#DolphinAsr-small Models

---

#FireRedAsr Series

#FireRedAsr-AED Chinese-English Model (v1)

#FireRedAsr2-AED Chinese-English Model (v2)

---

#Fun-ASR Series

Notes

• Model background: End-to-end speech recognition foundation model released by Tongyi Lab. Pre-trained on tens of millions of hours of real speech data, featuring strong contextual understanding and domain adaptability
• Features: All models are non-streaming, support punctuation, support timestamps. Support low-latency real-time transcription, with recognition accuracy reaching 93% in far-field, high-noise environments
• Version identifier meanings:
  • int8: INT8 quantized version, smaller size, faster inference, suitable for edge deployment
  • LLM: Large model enhanced version, stronger context understanding, suppresses recognition hallucinations
  • CTC: Lightweight classic CTC architecture version, lightweight inference
  • MLT: Multilingual general-purpose version, covers 31 languages
  • split-adaptor: Version with feature adaptation module deployed separately
• Language and capability notes:
  • Fun-ASR-Nano: Supports Chinese, English, Japanese; 7 dialects (Wu, Cantonese, Min, Hakka, Gan, Xiang, Jin); 26 regional accents (Henan, Shanxi, Hubei, Sichuan, Chongqing, Yunnan, Guizhou, Guangdong, Guangxi, Shaanxi, Hebei, Shandong, Anhui, Tianjin, Ningxia, Liaoning, Gansu, Hunan, Heilongjiang, Jilin, Inner Mongolia, Jiangsu, Zhejiang, Fujian, Jiangxi, Hainan); additionally supports lyrics recognition and rap speech recognition
  • Fun-ASR-MLT-Nano: Supports 31 languages total: Chinese, English, Cantonese, Japanese, Korean, Vietnamese, Indonesian, Thai, Malay, Filipino, Arabic, Hindi, Bulgarian, Croatian, Czech, Danish, Dutch, Estonian, Finnish, Greek, Hungarian, Irish, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Slovak, Slovenian, Swedish
• Domain advantages: Excellent performance in vertical fields such as education and finance, accurately recognizes domain-specific terminology, effectively suppresses hallucinations and language confusion

#Fun-ASR-Nano Models

#Fun-ASR-MLT-Nano Models

---

#FunASR Series

#Paraformer Chinese-English Models

#Paraformer Cantonese/Chinese/English Multilingual Models

#SeACo-Paraformer Hotword Customization Model

SeACoParaformer is a next-generation non-autoregressive speech recognition model with hotword customization, proposed by Alibaba Speech Lab. Compared to the previous CLAS-based hotword customization solution, SeACoParaformer decouples the hotword module from the ASR model and performs hotword boosting via posterior probability fusion, making the boosting process visible and controllable, while significantly improving hotword recall.

#SenseVoice Models

---

#K2TransducerAsr Series

#Streaming Models

#Non-streaming Models

---

#MedAsr Series

Notes

• Model architecture: Based on Conformer, a medical-domain speech recognition model released by Google Health
• Application scenarios: Suitable for radiology dictation, doctor-patient dialogue, medical transcription, etc.
• Supported languages: English only (primarily American English)
• Model characteristics: Pre-trained on approximately 5,000 hours of medical speech data, strong recognition of medical terminology. Performance on non-standard drug names and structured data such as dates/times may vary, suitable for fine-tuning to adapt to specific business scenarios

---

#moonshine Series

#moonshine-tiny Models

#moonshine-base Models

---

#WeNet Series

#Streaming Models

#Non-streaming Models

---

#Whisper Series

Notes

1. Models with -kv suffix have KV Cache inference acceleration enabled
2. All models support punctuation and timestamps. Output paragraph-level timestamps by default, can enable word-level timestamps via parameters
3. Language coverage:
   • Standard multilingual versions (tiny/small/medium/large-v1/large-v2): Support 99 languages (including Chinese, Cantonese, English, Japanese, Korean, Russian, Arabic, Vietnamese, Ukrainian, and other major world languages)
   • large-v3 / large-v3-turbo series: Extend low-resource languages beyond the 99, total approximately 106 languages. New additions include Zulu (zu), Maori (mi), Swahili (sw), Hausa (ha), etc., with significantly improved language identification
   • Full language list and codes:

af(Afrikaans), am(Amharic), ar(Arabic), as(Assamese), az(Azerbaijani), 
ba(Bashkir), be(Belarusian), bg(Bulgarian), bn(Bengali), bo(Tibetan), br(Breton), bs(Bosnian), 
ca(Catalan), cs(Czech), cy(Welsh), 
da(Danish), de(German), 
el(Greek), en(English), es(Spanish), et(Estonian), eu(Basque), 
fa(Persian), fi(Finnish), fo(Faroese), fr(French), 
ga(Irish), gl(Galician), gu(Gujarati), 
ha(Hausa), haw(Hawaiian), he(Hebrew), hi(Hindi), hr(Croatian), hu(Hungarian), hy(Armenian), 
id(Indonesian), ig(Igbo), is(Icelandic), it(Italian), 
ja(Japanese), jv(Javanese), 
ka(Georgian), kk(Kazakh), km(Khmer), kn(Kannada), ko(Korean), ku(Kurdish), ky(Kyrgyz), 
la(Latin), lb(Luxembourgish), lg(Ganda), lt(Lithuanian), lv(Latvian), 
mai(Maithili), mg(Malagasy), mi(Maori), mk(Macedonian), ml(Malayalam), mn(Mongolian), mr(Marathi), ms(Malay), mt(Maltese), my(Burmese), 
ne(Nepali), nl(Dutch), no(Norwegian), nso(Northern Sotho), ny(Chichewa), 
oc(Occitan), om(Oromo), or(Odia), 
pa(Punjabi), pl(Polish), ps(Pashto), pt(Portuguese), 
ro(Romanian), ru(Russian), rw(Kinyarwanda), 
sa(Sanskrit), sd(Sindhi), si(Sinhala), sk(Slovak), sl(Slovenian), sm(Samoan), sn(Shona), so(Somali), sq(Albanian), sr(Serbian), ss(Swati), st(Southern Sotho), su(Sundanese), sv(Swedish), sw(Swahili), 
ta(Tamil), te(Telugu), tg(Tajik), th(Thai), ti(Tigrinya), tk(Turkmen), tl(Tagalog), tn(Tswana), to(Tongan), tr(Turkish), ts(Tsonga), tt(Tatar), tw(Twi), 
ug(Uyghur), uk(Ukrainian), ur(Urdu), uz(Uzbek), 
ve(Venda), vi(Vietnamese), vo(Volapük), 
wa(Walloon), wo(Wolof), 
xh(Xhosa), 
yi(Yiddish), yo(Yoruba), 
zh(Chinese), yue(Cantonese), zu(Zulu)

• Language code short form:

af, am, ar, as, az,
ba, be, bg, bn, bo, br, bs,
ca, cs, cy,
da, de, el, en, es, et, eu,
fa, fi, fo, fr, ga, gl, gu,
ha, haw, he, hi, hr, hu, hy,
id, ig, is, it,
ja, jv,
ka, kk, km, kn, ko, ku, ky,
la, lb, lg, lt, lv,
mai, mg, mi, mk, ml, mn, mr, ms, mt, my,
ne, nl, no, nso, ny,
oc, om, or,
pa, pl, ps, pt,
ro, ru, rw,
sa, sd, si, sk, sl, sm, sn, so, sq, sr, ss, st, su, sv, sw,
ta, te, tg, th, ti, tk, tl, tn, to, tr, ts, tt, tw,
ug, uk, ur, uz,
ve, vi, vo,
wa, wo, xh,
yi, yo,
zh, yue, zu

#whisper-tiny Models

#whisper-small Models

#whisper-medium Models

#whisper-large Models

#Distil-Whisper Models

---

General Notes

• int8 = quantized version, smaller size, faster speed
• kv / selfcrosskv / selfcrosskvstack / opt = inference optimization versions
• Some models provide HuggingFace or GitHub sources; see each table