AI Tennis Radio

2023 prototype · Stats Perform internship

Opta point-by-point data went in, GPT-4 wrote the call, and stock synthetic voices read it aloud. An audio-first tennis broadcast, demoed over the 2023 Australian Open women’s final.

Opta Ace AI, in-game call Rybakina serving at 4–3, first set · two excerpts, 1:17 · AWS Polly voice “Matthew”

Side A

Four segments of one broadcast, voiced by a two-host desk, Arthur and Althea (AWS Polly stock voices “Arthur” and “Amy”). Every number they say came from Stats Perform data fed into the prompt.

  1. 1

    Pre-match report

    The hosts open the show from season and career stats before a ball is struck.

    0:00 / 0:39
    Transcript

    Arthur Welcome to the Stats Perform Pre-Match Report! I’m Arthur, joined by my esteemed co-host, Althea. We’re just moments away from the Australian Open Women’s Singles Final 2023, featuring Elena Rybakina and Aryna Sabalenka. What a matchup we have today, eh, Althea?

    Althea Absolutely, Arthur! We’re looking at two formidable players, each with her own distinctive strengths and achievements. Let’s start with Elena Rybakina. Hailing from Kazakhstan, Elena has had an impressive tour year in 2022, participating in 23 tournaments, winning 37 matches, and finishing as runner-up twice.

  2. 2

    Live point: break point at 15–40

    Called serve by serve from the point feed, over the match sound, until Sabalenka breaks for 4–4.

    0:00 / 0:43
    Transcript

    Althea A steady first serve from Rybakina this time, causing Sabalenka to hit a return forehand error. It’s now fifteen–forty, but Sabalenka still has a break point opportunity.

    Arthur Absolutely, Althea. Rybakina must dig deep here. She serves again, and we’re into a 6-shot rally. Sabalenka, though, comes out on top with a pinpoint backhand passing shot. And that’s the game. Sabalenka breaks, and we’re now at 4–4.

  3. 3

    Momentum match-up

    At the changeover: win probability then and now, and a momentum score explained for casual fans.

    0:00 / 0:59
    Transcript

    Althea Absolutely, Arthur. Let’s start with the pre-match win probabilities. Rybakina, the underdog, started with a 46.15% chance of winning. But as this match has unfolded, she’s stepped up her game, and her win probability is now at 60.65%.

    Arthur Contrast that with Sabalenka. She began the match with a slight advantage at 53.85%, but the tide has turned and her current win probability has dropped to 39.35%. Tennis, as we know, can be a game of shifting fortunes.

    Althea Indeed, Arthur. It’s also worth noting the momentum score here, which stands at 5.5 in Rybakina’s favor. Now, this might not mean much to casual fans, but it’s a significant figure calculated by Stats Perform. Anything over three points of momentum represents a considerable swing in favor of a player. Rybakina has certainly been setting the pace.

  4. 4

    Post-match breakdown

    The sign-off as Sabalenka takes her first Grand Slam title.

    0:00 / 0:30
    Transcript

    Arthur Good evening folks, I’m Arthur, and I’m joined by Althea to break down what was a thrilling Australian Open Women’s Singles 2023 Final. Aryna Sabalenka of Belarus has just claimed her first Grand Slam title, winning over Elena Rybakina of Kazakhstan.

    Althea Yes, Arthur, Sabalenka definitely played her heart out tonight. And for our audience, remember, we’re diving into our Stats Perform Breakdown, where we analyze key match statistics and explore the most crucial moments in the game.

How a point became a call

Each segment ran the same chain. It worked segment by segment on a finished match, not live, which was enough to test whether data-grounded AI commentary could hold a listener.

  1. Opta point feed

    Every point: server, score, serve, rally length, how it ended. Plus Stats Perform models: leverage, clutch points, momentum, live win probability.

  2. Prompt

    Python rewrote each point as a plain sentence, then wrapped the game in a commentator brief: who is talking, for which audience, what to explain.

  3. GPT-4

    Wrote the script, told to stay factual to the point data and explain the stakes: break points, second-serve pressure, clutch points.

  4. TTS voice

    Lines split by speaker and read by AWS Polly stock voices; Play.ht was tested alongside. Scores rewritten for the ear: “love–30”, not “0–30”.

  5. Broadcast

    Audio timed into the gaps between points on the match footage, in four formats: pre-match, in-game, changeover, post-match.

The prompt

An excerpt of the brief behind the Opta Ace call in the video, as written in the project notebook in summer 2023. The point lines underneath were generated from the feed.

"Imagine you are one tennis commentators broadcasting a live match."
- Prioritize sharing the game score, serve outcome, and how the point ends
  such that a even a non-tennis fan can clearly understand what is happening
- the point by point data is in chronological order and separated by point,
  do not make anything up in regards to the point information

commentator descriptions:
name "Opta Ace"
- Act as informational commentator who carefully explains in-depth
  statistics to help fans understand the player's strategy …
- Announce important moment such as Clutch Point, Break Point while
  loosely explaining them as a broadcaster would

Clutch point Definition: A clutch point is defined as any point with a
leverage of 10% or more – those which have a significant impact on the
final result of a match.

Point by Point information below:
New Game: Rybakina is now serving at 4 - 3 in set 1

Game Score: 0 - 0, Rybakina serving
Rybakina hits a First Serve, Sabalenka wins the point after a long rally
of 8 shots with a Forehand Winner

Game Score: 0 - 15, Rybakina serving
Rybakina hits a Second Serve, Sabalenka wins the point when Rybakina hits
a Backhand error
…