alakxender/dhivehi-conversations-turn

Dhivehi Conversations (Turn-Based)

This is an experimental synthetic dataset of turn-based Dhivehi conversations created for testing and fine-tuning dialogue models, text-to-speech (TTS), and multi-turn speaker-aware systems.

This dataset is artificially constructed and not based on real conversations. It is intended for research experimentation only and may not always produce contextually accurate results.

Dataset Source

Derived from alakxender/voice-synthetic, this dataset simulates a conversation structure by aligning synthetic Dhivehi sentences with turn and speaker information.

Dataset Structure

Each entry in the dataset includes:

• audio: Path or reference to the audio file for the utterance.
• texts: A list of Dhivehi utterances simulating a multi-turn conversation.
• speaker_ids: Identifiers for the speaker of each utterance:
  • 0: Female
  • 1: Male
• turn_ids: Numerical turn index (0, 1, 2...) indicating the dialogue order.
• conversation_id: A unique identifier linking all turns of a single conversation.

Example

{
  "audio": "path/to/audio.wav",
  "texts": [
    "ބައެއް ގޭގެ ދުވި ބުރިޔަށް ހަދައިފައިވާ ކަމަށްވެސްވެ",
    "އަދި އޭނާ އެކަމުގެ އުޖޫރައަކަށް ދިއްކޮށްލި ފަންސާސް ރުފިޔާގެ ނޫޓުގައި މީނާއަށް ހިފިއްޖެ"
  ],
  "speaker_ids": [1, 0],
  "turn_ids": [0, 1],
  "conversation_id": "conv_001"
}

Use Cases

• Turn-based dialogue modeling
• Speaker-conditioned TTS generation
• Evaluation of ASR → NLP → TTS pipelines
• Dhivehi multi-speaker conversational systems

Notes

• This dataset is meant for experimentation, prototyping, and non-production research tasks.
• Utterances were crafted to simulate human-like conversations, but are not based on actual dialogues or transcriptions.

Speaker Guide

Speaker ID	Voice Type
0	Female voice
1	Male voice