by DatapointAI
A gated Datapoint AI dataset of 315,000 blind pairwise human votes comparing 15 text-to-speech models on 300 English voice-agent prompts, with all 4,500 audio samples included.
DatasetFree
Step 1: Sign in to Hugging Face and accept the dataset's access conditions.
Step 2: Read the dataset structure section to choose a config: audio, pairs, responses, prompts or models.
Step 3: Load the config you need with the Hugging Face datasets library.
Step 4: Join audio and comparison rows using the audio keys.
Step 5: Use raw votes or the trust-weighted totals to train or evaluate a preference model.
The repository is public, but it is gated. You must agree to share your Hugging Face information and accept the conditions to access the files.
CC-BY-4.0 with attribution to Datapoint AI. You also agree not to re-identify annotators, and audio usage rights follow each model provider's terms.
315,000 eligible votes comparing 15 text-to-speech models over 300 prompts.
Eight voice-agent categories, including empathy and de-escalation, greetings and brand delivery, and instructions and troubleshooting.