Formosa Speech Recognition Challenge 2025 - Hakka ASR II

TAT_s2st_benchmark

Meta TAT_s2st_benchmark Project

Hokkien <-> English Speech to Speech Translation Benchmark

Description

This is the first (to our knowledge) Hokkien-English parallel speech dataset that is available for benchmarking Hokkien<->English speech to speech translation systems.
The dataset was created from TAT-Vol1-eval-lavalier (dev) and TAT-Vol1-test-lavalier (test), which contained audio recordings and transcripts in Taiwanese Hokkien.
Meta AI created the dataset by first concatenating neighboring sentences to form longer utterances, translating the Hokkien text transcriptions into English via Hokkien-English bilinguals, and recording the English translations with human voices.

Sample

ID: TAT-Vol1-test_0014_3.10_IUM001_concat

src_text_hanlo_tai

閣對台灣尾行到台灣頭，一步一跤印，用上謙卑的態度，去親近這塊晟養家己的土地，看著綠色的山脈、彎來斡去的河流、熟似的田園、純樸的鄉鎮、迷人的海岸、鄉親親切的笑容......，心肝頭定著有滿滿的感觸佮感動。

src_text_tailo:

koh uì Tâi-uân bué kiânn kàu Tâi-uân thâu, tsi̍t pōo tsi̍t kha-ìn, iōng siōng khiam-pi ê thāi-tōo, khì tshin-kūn tsit tè tshiânn-ióng ka-tī ê thóo-tē, khuànn-tio̍h li̍k-sik ê suann-me̍h, uan-lâi-uat-khì ê hô-liû, si̍k-sāi ê tshân-hn̂g, sûn-phok ê hiong-tìn, bê-lâng ê hái-huānn, hiong-tshin tshin-tshiat ê tshiò-iông......, sim-kuann-thâu tiānn-tio̍h ū buán-buán ê kám-tshiok kah kám-tōng.

src_text_tailo_number_tone:

koh4 ui3 tai5-uan5 bue2 kiann5 kau3 tai5-uan5 thau5, tsit8 poo7 tsit8 kha1-in3, iong7 siong7 khiam1-pi1 e5 thai7-too7, khi3 tshin1-kun7 tsit4 te3 tshiann5-iong2 ka1-ti7 e5 thoo2-te7, khuann3-tioh8 lik8-sik4 e5 suann1-meh8,uan1-lai5-uat4-khi3 e5 ho5-liu5,sik8-sai7 e5 tshan5-hng5, sun5-phok4 e5 hiong1-tin3,be5-lang5 e5 hai2-huann7,hiong1-tshin1 tshin1-tshiat4 e5 tshio3-iong5......, sim1-kuann1-thau5 tiann7-tioh8 u7 buan2-buan2 e5 kam2-tshiok4 kah4 kam2-tong7.

english:

I walked from the southernmost tip of Taiwan to the northernmost tip of Taiwan, one step at a time with a humble attitude, getting closer to this land where I was raised. Watching the green mountain ranges, curving rivers, familiar fields, simple towns, charming coasts, and the kind smiles of the villagers ...... I felt touched.

Dataset

License