Dual-Path RNN Separation Quiz

Answers and explanations appear as you take the quiz.

1. What problem does Dual-Path RNN primarily solve?
   • Modeling extremely long audio sequences that overwhelm a single RNN
   • Generating new music from text
   • Compressing audio files for storage
   • Detecting the language being spoken
2. What are the two 'paths' in a Dual-Path RNN?
   • Encoder and decoder
   • Intra-chunk (local) and inter-chunk (global) recurrence
   • Forward and backward audio playback
   • Training and inference paths
3. How is the long feature sequence prepared before the dual-path blocks?
   • It is converted to a single average vector
   • It is turned into text tokens
   • It is segmented into short overlapping chunks forming a 2D grid
   • It is downsampled to one sample
4. Into which existing framework was DPRNN dropped as a separator replacement?
   • A diffusion image model
   • A speech-to-text decoder only
   • A GAN-based vocoder
   • The Conv-TasNet encoder/decoder framework
5. Which later model kept DPRNN's dual-path structure but replaced the RNNs with Transformers?
   • SepFormer
   • WaveNet
   • BERT
   • Tacotron
6. Why does processing in chunks help training stability?
   • It removes the need for any loss function
   • Each RNN only sees a short sub-sequence, avoiding very long backprop chains
   • It eliminates all overlapping speakers automatically
   • It converts audio to images
7. How are processed overlapping chunks recombined into the output sequence?
   • By sorting them alphabetically
   • By averaging all chunks into one frame
   • By overlap-add across the chunk positions
   • By converting them into text tokens
8. On which benchmark did the DPRNN paper report its 20-times-smaller model comparison?
   • MUSDB18
   • ImageNet
   • LibriTTS
   • WSJ0-2mix