Speech Separation and the Cocktail Party Problem Quiz

Answers and explanations appear as you take the quiz.

1. What is the 'cocktail party problem'?
   • Focusing on one voice or sound source amid many competing ones
   • Choosing the right drink at an event
   • Measuring the loudness of a room
   • Compressing party music for streaming
2. What is the output of a speech separation system given a mixed recording of several speakers?
   • A single louder version of the mixture
   • A separate, isolated audio track for each speaker
   • A text summary only
   • A compressed MP3 of the whole room
3. What does Conv-TasNet do differently from many earlier separation methods?
   • It requires at least eight microphones
   • It only works on pre-recorded music
   • It operates directly in the time domain on raw waveform samples instead of spectrograms
   • It translates speech before separating it
4. How do many separation networks isolate an individual voice from the mixture?
   • By increasing the sample rate
   • By deleting all frequencies above 1 kHz
   • By converting audio to grayscale
   • By estimating a mask that selects each speaker's portion of the signal
5. What is 'target-speaker extraction'?
   • Pulling out one specific person using a short reference sample of their voice
   • Removing all human voices from a recording
   • Counting how many speakers are present
   • Boosting background music
6. Why can adding video help with speech separation?
   • Video reduces the file size of audio
   • Lip movements provide visual cues that help match sound to the correct speaker
   • Cameras cancel echo automatically
   • Video removes the need for any audio processing
7. What makes overlapping speech difficult for a separation system before it has analyzed the mixture?
   • The mixture contains no audio samples
   • Every speaker always occupies a separate frequency band
   • The number of speakers and which parts of the waveform belong to each speaker may be unknown
   • The recording already labels each word with its speaker
8. What cue can an early multi-microphone separation method use to distinguish speakers?
   • The spelling of each speaker's name
   • The color of the recording interface
   • The file's compression ratio alone
   • Spatial differences in how each voice reaches the microphone array