TugTalk Watch the demo

TugTalk · speaking practice for Apple Watch

Practise on your watch. Your assistant remembers.

Answer a question out loud on your Apple Watch. TugTalk counts your fillers, unsure words and restarts.

Watch the demo How it works

Each answer is saved to one record that Muse, Meta's assistant, reads on your phone and laptop, so the next chat starts from how you actually speak.

Tug, your sparring partner

Two fillers. Want another go?

Example answer

Um, so I think the main point is basically that we should ship the smaller version first.

  • 2 fillers
  • 1 unsure word

How one answer gets counted

About 15 seconds from Stop to a saved session.

  1. Watch app ready to record, with the goal Answer clearly, English, and a Record button

    Step 1 of 4

    Record on the watch

    Pick a goal, tap Record, answer, tap Stop.

  2. Um, so I think the main point is basically…

    Step 2 of 4

    Transcribed by Meta

    Meta's speech-to-text keeps “um” as a word, so it can be counted.

    • 2 fillers
    • 1 unsure word
    • 2 restarts

    Step 3 of 4

    Counted by code

    A fixed English or French word list finds each habit.

  4. Step 4 of 4

    Saved to one record

    Muse Spark Open adds one observation and one exercise. Muse reads it on your phone and laptop.

What one 14-second answer shows

Um, so I think the main point is basically that we should ship the smaller version first. Sorry, let me restart. The main point is ship the smaller version first because it lets us test with real users this week

Observation, from Muse Spark Open

“You met the goal after restarting, stating ‘The main point is ship the smaller version first’ followed by one supporting detail ‘it lets us test with real users this week’.”

2fillers“Um”, “basically”: 5 per 100 words
1unsure word“I think”
2restarts“Sorry”, “let me restart”
166words per minutea conversational pace

Recorded through the live service from a scripted clip read by a computer voice.

One practice record on three devices

Practise on the watch or the Mac. Muse on your phone reads the same record.

Watch app ready to record, with the goal Answer clearly and English selected
Apple Watch records
Speaking

You said

So, um, I think I could maybe lead the Q1 launch.

You said “maybe” lead the Q1 launch - are you asking me to let you lead it?

⌃⌥T Hold and answer

Mac asks a follow-up

Muse chat, example

How did my last practice go?

Your last capture for “Answer clearly” had 2 fillers, 1 unsure word and 2 restarts at 166 wpm. After the restart you put the main point first.

What should I practise next?

One answer in one take: “The main point is…”, then “because…”.

Phone remembers

A practice round on the Mac

  1. Tug waits in the top bar, around the camera notch.
  2. Hold ⌃⌥T and answer out loud.
  3. Tug, playing the other person, asks a follow-up that quotes your words and says it out loud.
The top bar opened into a card under the camera notch: Challenge, Round 1. What's the one fact that backs that up? You said “I think”. Buttons: Talk ⌃⌥T, Read this window.
  1. Listening, with the red record light
  2. Thinking
  3. Speaking
  4. Your turn

Sample round · the Mac top bar and its card

Muse Spark Open writes the question. Code keeps the quote only if you said it word for word. Wit.ai's Rebecca voice speaks it.

  • Reads only the window in front, only when you ask
  • Names, emails and numbers are hidden before window text leaves your Mac
Watch, after Tug speaks: You said “Um”. Say it again without that. Button: Record.
Sample round · Apple Watch

The same round on the watch

Tap Record and answer. Tug asks the follow-up through the watch speaker, and each answer joins the same record.

Answer 3 · watch simulator, computer voice

The main point is we ship the smaller version first. It lets us test with real users this week

“You said we ship the smaller version first, but that was very fast. Can you pause, then say it again much more slowly?”

Live round, Nov 2 · quote matched word for word

The model suggests. Code decides.

What Meta's models write, and what code checks before you see or hear it.

11 of 11quotes matched word for word12 live rounds, Nov 2

Model suggests

Code decides

  1. Meta speech-to-textWrites down the words, “um” included

    Counts fillers, unsure words and restarts

  2. Muse Spark OpenWrites the follow-up question

    Keeps a quote only if you said it word for word

  3. Muse Spark OpenNext release: assesses whether your answer addresses the request

    Next release: checks evidence and saves round state across devices

  4. MuseNext release: saves a drill with its source sessions

    Next release: recounts the drill’s numbers and refuses a wrong count

Before and after the restart

Um, so I think the main point is basically that we should ship the smaller version first.

  • 2 fillers
  • 1 unsure word

Sorry, let me restart.

  • 2 restarts

The main point is ship the smaller version first because it lets us test with real users this week

  • 0 fillers
  • 0 unsure words

The same 14-second answer, split at the restart. The restart phrase itself counts as 2 restarts, so the three parts add up to the counts above.

Watch the demo

Frequently asked questions

How are my speaking habits counted?

TugTalk turns your recording into text, then counts words and phrases such as “um” and “maybe” using a fixed English or French word list. The transcript can contain mistakes, so a count may be wrong.

Are my recordings saved?

Your audio is deleted after it has been turned into text, unless you turn on Keep my recordings. That setting is off by default. You can delete saved audio at any time. Your transcript and results stay in your account.

Does a lower count mean I’m improving?

One answer is not enough to tell. The counts describe that answer; they do not prove that your speaking has improved.

Do I need to use my camera?

No. It is off by default. If you turn it on, your Mac measures which way your head faces. It sends those measurements to TugTalk, without sending camera images. It does not track your eyes.

Coming next · in development

French speaking practice

A short spoken debrief in French becomes a few phrases to repair and reuse. Confirmed phrases export to Anki.

Said
« Avant le cours, j'ai allé à la bibliothèque. »
Repaired
« Avant le cours, je suis allé à la bibliothèque. »
  1. Suggested
  2. Confirmed
  3. Repaired
  4. Reused