Dataset · keithito
Classic English speech dataset LJ Speech
keithito/lj_speech
24 hours of single-speaker English studio speech with text transcripts for ASR and TTS.
What's inside
The dataset includes 13,100 short audio clips lasting from 1 to 10 seconds in .wav format, recorded by a single speaker reading excerpts from 7 nonfiction English books. Total recording duration is about 24 hours, with a text transcription provided for each clip. The dataset is in the public domain.
What you can do with it
- Training and fine-tuning single-speaker text-to-speech (TTS) models
- Training and testing English automatic speech recognition (ASR) systems
- Debugging audio pipelines and vocoders on clean speech data
- Tasks
- Automatic speech recognition, Text → speech, Text → audio
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 1.4K
- Likes
- 62
Models trained on it
- 11mlabs/indri-0.1-124m-ttsText → speech 982
- AuroraAI-Research/AuroraTTS-100KText → speech 461
- mradermacher/indri-0.1-124m-tts-GGUF 300
- mradermacher/indri-0.1-350m-tts-GGUF 262
- SupraLabs/SupraTTS-0.1-BetaText → speech 160
- shangeth/Wren-TTS-0.5B-multi-expressiveText → speech 155
- ibm-research/DAC.speech.v1.0 129
- muhamadul/xtts-v2-wolofText → speech 63
How to load it
from datasets import load_dataset
ds = load_dataset("keithito/lj_speech")