HKUSTAudio/Llasa_opensource_speech_data_160k_hours_tokenized

Update (2025-02-07): Our paper has been released! This script is for merging tokenized speech datasets stored in memmap format. The input datasets can be combined to form larger training datasets. import numpy as np import os def merge_memmap_datasets(dataset_dirs, output_dir): # Ensure the output directory exists os.makedirs(output_dir, exist_ok=True) # Dataset splits to be merged splits = ['train', 'val'] for split in splits: shapes = [] seq_len =… See the full description on the dataset page: https://huggingface.co/datasets/HKUSTAudio/Llasa_opensource_speech_data_160k_hours_tokenized.

<iframe src="https://silytics.si/embed/item/79ff2a78-fd6f-4469-8365-6a96cc2b1d80" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 18 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.