What Datapoint AI's 315K-Vote TTS Dataset Actually Tells You — and What Its Pricing Chart Can't
𝟯𝟭𝟱,𝟬𝟬𝟬 𝗵𝘂𝗺𝗮𝗻 𝗿𝗮𝘁𝗶𝗻𝗴𝘀 𝗼𝗳 𝗧𝗧𝗦 𝗺𝗼𝗱𝗲𝗹𝘀, 𝗳𝗿𝗲𝗲 𝘁𝗼 𝗱𝗼𝘄𝗻𝗹𝗼𝗮𝗱. Datapoint AI just opened the largest human audio preference dataset I've seen, built specifically around customer support. Not WER. Not MOS. Real people choosing which voice they'd rather hear on a support call. ━━━━━━━━━━━━━━━ 𝗪𝗵𝗮𝘁 𝗺𝗮𝗸𝗲𝘀 𝗶𝘁 𝘂𝘀𝗲𝗳𝘂𝗹 → 300K+ annotations from real annotators → 15 TTS models ranked head to head → 8 categories: IVR menus, empathy, escalations, refunds and more That category split is the part people will scroll past, and it matters most. An IVR menu voice and an escalation voice are different jobs. One needs to be legible at speed. The other needs to sound like it is not enjoying your complaint. A single quality score averages those into mush. If you build support agents, you have been guessing at this. Now there is data. ━━━━━━━━━━━━━━━ 𝗧𝗵𝗲 𝗮𝘅𝗶𝘀 𝘁𝗵𝗲 𝗰𝗵𝗮𝗿𝘁 𝗰𝗮𝗻𝗻𝗼𝘁 𝘀𝗵𝗼𝘄 The quality-cost frontier plots price in dollars per 1M characters. Every model on that axis is a rental. Per-character billing is the only cost model it can express. A model running on your own GPU has no per-character price at all. Its cost is hardware and electricity, and it drops toward zero as volume climbs. It cannot be plotted on that x-axis, so it mostly is not there. That is not a knock on the benchmark. It is a reminder that a leaderboard built on API pricing will always frame self-hosting as missing rather than as a different shape. ━━━━━━━━━━━━━━━ 𝗧𝗵𝗲 𝗿𝗲𝗮𝗹 𝗮𝘀𝘀𝗲𝘁 𝗶𝘀 𝘁𝗵𝗲 𝗱𝗮𝘁𝗮𝘀𝗲𝘁, 𝗻𝗼𝘁 𝘁𝗵𝗲 𝗿𝗮𝗻𝗸𝗶𝗻𝗴 The leaderboard tells you what 15 vendors scored last month. It is stale the moment someone ships a new checkpoint. 315K human preference pairs are something else entirely. → Re-rank on the categories your business actually runs → Train a reward model for automated TTS evaluation → Use it as preference data to align an open-weight model you own A scoreboard is a snapshot. Preference data is leverage. ━━━━━━━━━━━━━━━ Rankings and pricing above are Datapoint's own published numbers. I have not reproduced them. Neither has anyone else reposting the chart today. I build enterprise voice agents that run on customer infrastructure — TTS, telephony, escalation logic, no audio leaving the building.
315,000 human ratings of TTS models for customer support — real, and worth understanding beyond the headline.
Datapoint AI has published one of the largest human audio preference datasets built specifically for customer-support voice use cases: 315,000 pairwise votes comparing 15 TTS models across 300 prompts, split into eight practical categories covering things like empathy and de-escalation, greetings, troubleshooting instructions, and transactional readouts. Rather than a single quality score, votes are broken out by category — recognizing that a voice suited to reading account numbers legibly and a voice suited to de-escalating an angry customer are different jobs, and averaging them into one number would obscure that.
One limitation worth knowing: the accompanying pricing/quality comparisons, like similar industry benchmarks (e.g. Artificial Analysis), are typically expressed in dollars per million characters — a metric that only works for API-billed models. Self-hosted, open-weight models don't have a meaningful per-character price; their cost is hardware and electricity, which falls per unit as volume scales. That's not a flaw in the benchmark methodology, but it does mean self-hosted options will tend to look "missing" from cost charts built around API pricing, rather than genuinely absent from consideration.
The more durable asset here isn't the leaderboard snapshot, which will go stale as vendors ship new versions — it's the underlying 315,000 preference pairs themselves, which can be re-ranked for a specific business's actual usage categories, used to train a reward model, or applied as preference data to fine-tune an open-weight model.
Access to the dataset requires a free Hugging Face account and agreeing to its usage terms (it's licensed CC-BY-4.0, with the underlying generated audio separately governed by each TTS provider's own terms).