Gemini Flash 2.0 Speech: 94,512 Synthetic Audio Samples from Two Speakers

94,512 high-quality synthetic English speech audio files generated by Gemini Flash 2.0 via the Multimodal Live API. The dataset contains speech from two speakers, Puck (Male) and Kore (Female), totaling 284.31 hours with an average duration of 10.83 seconds per file. It was created by user shb777 and was trending in February 2025.

Use Cases

• Training or fine-tuning text-to-speech models based on high-quality synthetic speech data.
• Benchmarking speech synthesis quality using audio generated by a state-of-the-art multimodal API.
• Studying speaker characteristics and voice consistency across a large corpus of synthetic audio.
• Developing multi-speaker TTS systems based on the two distinct voices (male and female) provided.

Strengths

• Large scale with 94,512 audio files totaling 284.31 hours of speech.
• High-quality synthetic generation using Gemini Flash 2.0, a state-of-the-art model.
• Includes two distinct speaker voices (male and female), enabling multi-speaker research.
• Reportedly used in training notable models like Kokoro TTS and LLaSA 1B, indicating practical utility.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Data is entirely synthetic, which may limit applicability for tasks requiring natural human speech.
• Description metadata is limited; actual data quality and file formats require manual inspection after download.

Provenance

Source

Generated by Gemini Flash 2.0 via the Multimodal Live API.

Collection Method

Synthetic generation.

Time Range

null

Freshness

Last updated 2025-06-26 16:10:08.

Geography

null