Open-source the kushinada - hubert - base model, and it can be freely deployed for Japanese speech feature extraction!

Kushinada Hubert Base

Developed by imprt

Japanese speech feature extraction model pre-trained on 62,215 hours of Japanese TV broadcast audio data

JapaneseOpen Source License:Apache-2.0 #Japanese speech feature extraction #Large-scale pre-training #TV broadcast audio

Downloads 1,922

Release Time : 3/7/2025

Model Overview

This is a HuBERT base model specifically designed for Japanese speech, pre-trained on large-scale Japanese TV broadcast audio data, suitable for speech feature extraction tasks.

Model Features

Large-scale pre-training data

Pre-trained on 62,215 hours of Japanese TV broadcast audio data to ensure deep understanding of Japanese speech

Professional speech feature extraction

Optimized based on HuBERT architecture, specifically designed for extracting high-level feature representations of Japanese speech

TV broadcast domain adaptation

Training data comes from TV broadcasts, providing good adaptability to formal spoken language and broadcast speech

Model Capabilities

Japanese speech feature extraction

Voice activity detection

Speech representation learning

Use Cases

Speech processing

Speech recognition preprocessing

Used as a front-end feature extraction module for Japanese speech recognition systems

Speech content analysis

Used to analyze speech features in TV broadcast content

🚀 `imprt/kushinada-hubert-base`

This is a Japanese HuBERT Base model pre-trained using 62215 hours of audio extracted from large-scale Japanese TV broadcast audio data by voice activity detection. This model provides valuable features for speech-related tasks.

🚀 Quick Start

This is a Japanese HuBERT Base model pre-trained using 62215 hours of audio extracted from large-scale Japanese TV broadcast audio data by voice activity detection.
This model was trained using code from the official repository.

💻 Usage Examples

Basic Usage

import soundfile as sf
from transformers import AutoFeatureExtractor
model = "imprt/kushinada-hubert-base"
feature_extractor = AutoFeatureExtractor.from_pretrained(model)
audio_file="/path/to/16k_audio_file"
audio_input, sr = sf.read(audio_file)
feature_extractor(audio_input, sampling_rate=sr)

📚 Documentation

References

@article{journals/corr/abs-2106-07447,
  added-at = {2021-06-16T00:00:00.000+0200},
  author = {Hsu, Wei-Ning and Bolte, Benjamin and Tsai, Yao-Hung Hubert and Lakhotia, Kushal and Salakhutdinov, Ruslan and Mohamed, Abdelrahman},
  biburl = {https://www.bibsonomy.org/bibtex/2435bd8c9ac37a4eab204ded15e9f8918/dblp},
  ee = {https://arxiv.org/abs/2106.07447},
  interhash = {c85407653eddc9c9256c261afe8d6954},
  intrahash = {435bd8c9ac37a4eab204ded15e9f8918},
  journal = {CoRR},
  keywords = {dblp},
  timestamp = {2024-04-08T22:55:35.000+0200},
  title = {HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units.},
  url = {http://dblp.uni-trier.de/db/journals/corr/corr2106.html#abs-2106-07447},
  volume = {abs/2106.07447},
  year = 2021
}

📄 License

Apache License, Version 2.0

⚠️ Important Note

Please read Apache License, Version 2.0 before downloading this model.

Property	Details
Model Type	Japanese HuBERT Base model
Training Data	62215 hours of audio extracted from large-scale Japanese TV broadcast audio data by voice activity detection
Pipeline Tag	feature-extraction
Tags	hubert, speech
License	apache-2.0
Language	ja
Extra Gated Prompt	Please read Apache License, Version 2.0 before downloading this model.
Extra Gated Fields	Country: country, Affiliation: text, I agree ALL the statements in Apache License, Version 2: checkbox
Extra Gated Button Content	Acknowledge license

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご