Dataset · k2speech
60 hours of transcribed Uzbek speech
k2speech/FeruzaSpeech
A read-speech dataset in Uzbek with Cyrillic and Latin transcriptions for ASR and speech synthesis.
What's inside
Contains audio recordings of read speech and text transcriptions in Uzbek in both Cyrillic and Latin scripts. The collection includes 10K to 100K examples totaling 60 hours, recorded by a single native female speaker from Tashkent, with data partially formatted in TSV.
What you can do with it
- Training and fine-tuning Uzbek text-to-speech (TTS) models
- Testing and adapting automatic speech recognition (ASR) systems for Uzbek Cyrillic and Latin scripts
- Languages
- Uzbek
- Size
- 10K–100K rows
- Downloads this month
- 738
- Likes
- 8
- Access
- on request on Hugging Face
Models trained on it
- navai-uz/whisper-medium-uzbekAutomatic speech recognition 3.8K
- navai-uz/whisper-small-uzbekAutomatic speech recognition 1.1K
- navai-uz/whisper-base-uzbekAutomatic speech recognition 338
- Mannon/whisper-medium-uzbekAutomatic speech recognition 179
- Mannon/whisper-small-uzbekAutomatic speech recognition 137
- navai-uz/whisper-tiny-uzbekAutomatic speech recognition 74
- SamariddinS/uzbek-voiceAutomatic speech recognition 0
How to load it
from datasets import load_dataset
ds = load_dataset("k2speech/FeruzaSpeech")