Neural Audio Codecs Quiz

Answers and explanations appear as you take the quiz.

1. What two things does a neural audio codec primarily do?
   • Compress audio into discrete tokens and reconstruct it
   • Record and amplify audio
   • Translate and transcribe speech
   • Add reverb and equalization
2. Which quantization technique is central to codecs like SoundStream and EnCodec?
   • JPEG quantization
   • Residual Vector Quantization (RVQ)
   • Linear predictive coding
   • Fourier quantization
3. In Residual Vector Quantization, what does each successive codebook encode?
   • A completely separate audio channel
   • The text transcript of the audio
   • The residual error left by the previous codebooks
   • The sample rate metadata
4. Why are neural audio codecs important for generative audio models?
   • They label audio with genres automatically
   • They mix multiple tracks together
   • They convert audio to sheet music
   • They provide the discrete tokens that models like VALL-E and MusicGen generate
5. How does using more codebooks in a neural codec affect the output?
   • It increases bitrate but improves reconstruction quality
   • It lowers the bitrate and quality together
   • It changes the language of the speech
   • It has no effect on quality
6. What subjective comparison did the SoundStream paper report at 3 kbps?
   • At 24 kHz, SoundStream at 3 kbps outperformed Opus at 12 kbps in the reported evaluation
   • It matched every codec at every bitrate
   • It was tested only against MP3 at 320 kbps
   • It produced higher quality than all commercial audio codecs
7. What is the decoder’s role in a neural audio codec?
   • Choose which transcript is legally correct
   • Predict the speaker’s identity from text only
   • Reconstruct audio from the encoded discrete representation
   • Assign language labels to silence
8. What inference capability did the SoundStream paper report for its low-latency codec?
   • Only offline batch processing on a server cluster
   • It required a GPU for every audio frame
   • It could encode images but not audio
   • Streamable real-time inference on a smartphone CPU