oto

@otodotearth

Speech Dataset Company

752Followers

5Following

50Posts total

16.3KViews on collected posts

Against accounts of the same size

7 posts from the last 90 days, next to the under 10K follower range. below its peers on both reach and engagement.

Median views430this account2 112median for under 10K

Reach, %57.18%this account157.00%median for under 10K

Engagement, %0.99%this account1.62%median for under 10K

Metric	This account	Median for under 10K	Ratio
Median views per post	430	2 112	0.20×
Reach (views ÷ followers)	57.18%	157.00%	0.36×
Engagement rate	0.99%	1.62%	0.62×

Others in this range → Compare with another account → How these benchmarks are built →

Growth & engagement

How the posts we collected actually performed: views and reaction rate post by post, what the audience did with them, and where the follower count goes.

Views per post

80426 Aug

473

13.6K

430

409

18027 Aug

Last 7 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

1.00%26 Aug

0.85%

1.33%

0.93%

2.20%

0.98%

2.78%27 Aug

Reactions — likes, reposts, replies and quotes — divided by views. Median for under 10K accounts is 1.62%.

What the audience does

Likes53.8%193 in total

Reposts2.5%9 in total

Replies3.1%11 in total

Quotes0.6%2 in total

Bookmarks40.1%144 in total

Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.

The follower curve appears once this account has two daily snapshots — we take one a day, and this one is on its first.

Latest posts

@otodotearth Great work Oto team! 180 views · 4 likes · 0 reposts · 1 replies 27 Aug 2026 · Open on X →

One change did that: the training data. Ours cut false interruptions from 63.0% to 33.0%. The same fine-tune also lifted the model on a task-oriented benchmark it never trained on, 35.2% to 21.6%. https://t.co/TVwaBBzzMq 409 views · 3 likes · 0 reposts · 1 replies 26 Aug 2026 · Open on X →

That is what we build datasets for: voice AI that has to work in natural conditions, and the models that come after it. Have a look at what we collect. https://t.co/1RIvYgURFo 409 views · 8 likes · 1 reposts · 0 replies 26 Aug 2026 · Open on X →

On that evaluation, every system we tested did worse than its task-oriented numbers suggest. Five current detectors lost 23.9 to 34.9 points at a 300ms budget. Voice AI needs models that hold up in the wild, not only in ideal conditions. https://t.co/ngpRPSfstz 430 views · 3 likes · 0 reposts · 1 replies 26 Aug 2026 · Open on X →

We fine-tuned a public end-of-turn model on full-duplex conversation we collected ourselves, with turn-taking labels on every pause. False interruptions fell from 63.0% to 33.0%. https://t.co/EEeJbsXN2o 13.6K views · 165 likes · 8 reposts · 6 replies 26 Aug 2026 · Open on X →

So we scored the hard part directly. Every internal pause in a long turn becomes one decision: end of turn, or still holding the floor. 11,568 decisions from 84 long conversations, all human reviewed. Each system decided from audio that stopped 200ms after the silence began. 473 views · 3 likes · 0 reposts · 1 replies 26 Aug 2026 · Open on X →

Existing detectors do reasonably well in task-oriented scenarios. Real conversation is different. It is spontaneous, and where one turn ends is often unclear. False cutoffs climb in exactly those moments. Training on our labelled conversation data moved that number a long way. 804 views · 7 likes · 0 reposts · 1 replies 26 Aug 2026 · Open on X →