Audio · Text · Multimodal / AI & Machine Learning
Conversational Text-to-Speech (TTS) Dataset | 20,000 Hours of 48kHz Voice Data for Text-to-Speech, Voice Cloning & Speech AI
Advertised by providerLicensing rights not independently verifiedPrice on request
Overview
Conversational Text-to-Speech (TTS) Dataset | 20,000 Hours of 48kHz Voice Data for Text-to-Speech, Voice Cloning & Speech AI is a multimodal dataset. The provider states a volume of 20,000 hours. Typical applications include speech recognition, speech synthesis (TTS) and conversational AI. Relevant sectors: AI & Machine Learning. Pricing is not published and is provided on request.
Primary applications
What the listing offers
- Stated volume of 20k hours.
The TTS Voice Dataset is a 20,000-hour collection of high-fidelity 48kHz conversational audio recorded globally across numerous languages, regions, accents, and speaker profiles. The dataset is designed for text-to-speech, voice cloning, speech synthesis, conversational AI, speaker modeling, multilingual speech training, and voice agent development. The collection emphasizes natural spoken dialogue rather than overly scripted or robotic recordings, making it useful for building expressive, human-like voices that can support real conversational products. Audio coverage includes high-demand commercial languages, major regional languages, and underrepresented long-tail languages that are difficult to source at scale. The dataset includes broad coverage across popular global languages such as English, Spanish, Mandarin Chinese, Hindi, Arabic, Portuguese, French, German, Japanese, Korean, Italian, Russian, Indonesian, Vietnamese, Thai, Turkish, Bengali, Urdu, Tamil, Telugu, Marathi, Dutch, Polish, Malay, and Swahili. Additional long-tail and underrepresented language coverage includes Tagalog, Uyghur, Hmong, Cebuano, Tamazight, Tibetan, Sepedi, Guaraní, Madurese, and Bouyei. The dataset is structured to balance commercial usefulness with linguistic diversity. Approximately 30% of the dataset, or 6,000 hours, covers English and major English accents such as American, British, Canadian, and Australian English. Approximately 35%, or 7,000 hours, covers major commercial global languag
Related datasets
More in Multimodal & Agent Training →Audio · Text · MultimodalAdvertised
Spontaneous Conversational Sundanese Speech Dataset — Consented, Transcribed (West Java, Indonesia)
Real Sundanese, spoken the way it is actually lived.
- Language
- —
- Volume
- —
- Domain
- AI & Machine Learning +1
- Use case
- Speech recognition
USD 81
More speech & audio datasets
All 1,727 speech & audio datasets →- Corsican live conversational audio, channel-separated
- Corsican live conversational audio, single-channel
- Coworking Space Conversation Audio Dataset (Global, English)
- Credit Risk Panel Audio Dataset (Europe, English)
- Croatian conversational audio, channel-separated speakers
- Croatian live conversational audio, channel-separated
- Croatian live conversational audio, single-channel
- Crunk hip-hop music tracks
- Custom AI Training Data | Audio, Video, Embodied AI, 3D & Text — Any Modality, Volume | Made-to-Order, Scoped Per Project | Machine Learning (ML) Data
- Custom On-Site Speech & Audio Data Collection