Update zoengjyutgaai_saamgwokjinji.py

[indiejoseph]

Since the dataset on Huggingface has been updated, the split "train" has been removed and replaced by "saamgwokjinji", "seoiwuzyun".

[AlienKevin]

Thanks! Can you add the mouzaakdung split as well?

[indiejoseph]

Done, but I tried run it, it contains much more audio samples then before, and takes too long time to run it, I think it is unnecessarily to through all samples

[AlienKevin]

Sure, maybe we can select a random subset from each dataset?

[indiejoseph]

The dataset is keep updating, it is hard for benchmarking when things changing, see would @laubonghaudoi could create a test split with 1k samples in the repo, or we fork one.

[laubonghaudoi]

如果全部都跑一次嘅話要幾多時間？

[indiejoseph]

H100 用 SenseVoice batch size 64 好快，大約45分鐘，但 whisper v3 好慢，我有用 custom processor 所以冇用 pipeline 去跑 batch, 所以唔可以作準, 遲啲用 pipeline 跑一次再報告。

[laubonghaudoi]

唔該晒，原來 H100都要咁耐。呢個PR可唔可以順便改埋個README？而家仲寫住

[Storytelling] Zoeng Jyutgaai's Romance of the Three Kingdomgs: 1,402 utterances by Cantonese storyteller Zoeng Jyutgaai (張悦楷) on the classical work "Romance of the three kingdoms".

個數據集名已經改成 CanCLID/zoengjyutgaai而且都唔止咁少 utterance