ReazonSpeech

by Reazon Holdings

Japanese ASR corpus — 35k hours of TV recordings with captions.

TL;DR

Japanese ASR corpus — 35k hours of TV recordings with captions.

Best for large-scale Japanese ASR pretraining. Pricing: research-only.

Category
Open source
License
—
Stars
—
Last push
—
Pricing
research-only
Platforms
HuggingFace

What it is

ReazonSpeech is up to 35k hours of Japanese TV speech with closed-caption alignment. Subsets: tiny / small / medium / large / all. License: CDLA-Sharing-1.0 (with broadcast caveats).

Best for: Large-scale Japanese ASR pretraining.
Watch out for: CDLA-Sharing-1.0 license · Japanese broadcast TV → strict source rights · research-friendly but copyright complex. Cite: Yin et al., NLP 2023.

Install / use

from datasets import load_dataset; ds = load_dataset('reazon-research/reazonspeech', 'small')

Features

Speaker diarizationNo
Word-level timestampsNo
Streaming / real-timeNo
Languages supported1
HIPAA eligibleNo

ReazonSpeech vs Whipscribe

FeatureReazonSpeechWhipscribe
CategoryOpen sourceTranscription APIs
Pricingresearch-onlyfree beta
Speaker diarizationNoYes
Word timestampsNoYes
StreamingNoNo
Languages199
PlatformsHuggingFaceWeb, API, MCP

Alternatives to ReazonSpeech

OpenAI Whisper
OpenAI

The reference open-source multilingual ASR model from OpenAI.

OSS · MIT ★ 98.1k
whisper.cpp
Georgi Gerganov

C/C++ port of Whisper — runs on anything, from a Raspberry Pi to Apple Silicon.

OSS · MIT ★ 48.8k
faster-whisper
SYSTRAN

4× faster than reference Whisper using CTranslate2 — production sweet spot.

OSS · MIT ★ 22.3k

Whipscribe is a managed faster-whisper + whisperX service. If you want transcripts without running infrastructure, paste a URL or drop a file in the form below — you'll have a transcript in seconds.