README.md

metadata

dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: text
      dtype: string
    - name: source
      dtype: string
    - name: sample_rate
      dtype: int64
    - name: speaker
      dtype: string
  splits:
    - name: train
      num_bytes: 2942869578
      num_examples: 13000
  download_size: 3303296950
  dataset_size: 2942869578
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: cc
task_categories:
  - text-to-speech
  - automatic-speech-recognition
language:
  - tr
size_categories:
  - 10K<n<100K

Synthetic Turkish TTS Data

This dataset was created by generating synthetic Turkish text across multiple speech scenarios. The text was produced in the following domains: finance_master, cs_master, parcel_delivery, ecommerce, telecom, isp_support, technical_support, subscription, insurance, health_appointments, public_services, education_registration, and daily_speech.

These synthetic texts were then synthesized with a high-quality Turkish TTS model. The dataset is intended to be used as synthetic training data for future Turkish TTS model training.

Dataset Summary

• Total number of samples: 13,000
• Total number of speakers: 4
• Total duration: 28 hours 50 minutes 13 seconds

Speaker Distribution

• ali: 6 hours 50 minutes 57 seconds (3250 wav files)
• zeynep: 6 hours 50 minutes 35 seconds (3250 wav files)
• leyla: 7 hours 10 minutes 55 seconds (3250 wav files)
• alev: 7 hours 57 minutes 46 seconds (3250 wav files)

Columns

• audio: audio file
• text: corresponding transcription
• source: scenario/domain from which the sample was generated
• sample_rate: audio sampling rate
• speaker: speaker name

License

This dataset is released under the CC BY 4.0 license. You may use, share, and adapt the dataset provided that proper attribution is given.