Comparing 11 Japanese transcription models with free GPUs: the 1st place in reading aloud ranked 8th for broadcast audio
To identify the best public model for Japanese transcription, 11 models were compared under the same conditions using Google Colab's free T4, with results published on Hugging Face.
Regarding which public model should be used for Japanese transcription, there is currently no consolidated comparison available. While the Open ASR Leaderboard on Hugging Face provides English-centric metrics, Japanese is not included in the multilingual column. Although there are some standalone articles on corporate blogs regarding Japanese comparisons, the conditions vary for each.
To address this, 11 models were measured under the same conditions. The measurement used the free T4 from Google Colab, incurring no costs. The results are placed on Hugging Face and include the output for each utterance. The model that ranked 1st in reading aloud ranked 8th for broadcast audio.
This item is a summary. The full, checkable version is at the original source:
Read the original at Zenn (Hugging Face) ↗