Glow-TTS Monotonic Alignment Quiz

Answers and explanations appear as you take the quiz.

1. What problem with attention-based TTS does Glow-TTS aim to fix?
   • Skipping or repeating words on long sentences
   • Audio that is too quiet
   • Inability to use a GPU
   • Lack of a text encoder
2. What does MAS stand for in Glow-TTS?
   • Mel Audio Synthesis
   • Monotonic Alignment Search
   • Multi-Attention System
   • Modular Acoustic Sampler
3. Why is the alignment required to be monotonic?
   • To make the audio louder
   • To reduce file size
   • Because text is read in left-to-right order over time
   • To support multiple speakers
4. What type of decoder does Glow-TTS use to model spectrograms?
   • A recurrent attention network
   • A convolutional classifier
   • A transformer language model
   • A normalizing flow
5. At inference time, how does Glow-TTS decide how long each text token lasts?
   • A duration predictor outputs frame counts per token
   • It runs MAS again on the audio
   • It guesses randomly
   • It uses a microphone
6. What is a practical advantage of Glow-TTS being non-autoregressive?
   • It needs no text input
   • It can generate all frames in parallel for fast synthesis
   • It cannot control pitch
   • It only works offline
7. What does the surjective-alignment assumption mean in Glow-TTS?
   • Each speech frame maps to multiple languages
   • Some text tokens must be skipped
   • Every text token maps to at least one speech frame
   • The output audio has no frame sequence
8. How does Glow-TTS find a likely alignment during training?
   • A human marks every phoneme-frame pair by hand
   • It randomly permutes the text tokens
   • A waveform classifier chooses a speaker identity
   • Dynamic programming searches for the most probable valid monotonic path