Word timings

let result = try await tts.generate("Highlight each word as it plays.")

for timing in result.wordTimings {
    print(timing.wordIndex, timing.word, timing.startTime, timing.endTime)
}

Each KittenWordTiming contains:

Field	Meaning
wordIndex	Zero-based word index within the generated result
word	Original display text for the word
startTime	Start in seconds
endTime	End in seconds

Timings are derived from the model's predicted phoneme durations, not a second forced-alignment pass.

UI synchronization

Start the elapsed-time clock when playback actually starts. For native speaker output through speak(), the current public API does not expose a playback-start callback; generate first and use your own AVAudioEngine or player when precise synchronized UI is required.

Limits

• Timings can be empty when duration data is unavailable.
• Multi-chunk text can omit timing data.
• Punctuation and normalization can affect word mapping.
• Generate sentences or short paragraphs for best highlighting behavior.

generateStreaming() produces a separate timing array for each sentence result.