Speech Denoising with RNNoise Quiz

Answers and explanations appear as you take the quiz.

1. What does RNNoise primarily predict for each short frame of audio?
   • A gain (0 to 1) for each of about 22 frequency bands
   • The transcript of the speech
   • The identity of the speaker
   • The language being spoken
2. What type of neural network is at the core of RNNoise?
   • A large transformer
   • A small recurrent network using GRUs
   • A convolutional image classifier
   • A diffusion model
3. Why does RNNoise use perceptual frequency bands instead of raw spectral bins?
   • To make the network huge and accurate
   • To translate audio into text
   • To keep the model tiny and reduce musical-noise artifacts
   • To increase the sample rate
4. Roughly how large is the RNNoise model?
   • About 1 billion parameters
   • About 50 million parameters
   • It has no learnable parameters
   • About 85,000 weights
5. What is the role of the pitch filter in RNNoise?
   • To remove residual noise between harmonics of voiced speech
   • To raise the speaker's pitch
   • To detect the speaker's age
   • To compress the file
6. Which popular application includes RNNoise as a built-in noise-suppression option?
   • Microsoft Excel
   • OBS Studio
   • Adobe Photoshop
   • A web browser's spellchecker
7. Which hand-crafted features does RNNoise feed to its recurrent network?
   • Image color histograms and edge maps
   • Wordpiece embeddings and attention masks
   • Band energies, pitch period, and pitch correlation
   • Speaker names and calendar timestamps
8. What does RNNoise’s separate voice-activity output help the system do?
   • Identify the speaker by name
   • Choose the bitrate of an audio codec
   • Translate the speech into another language
   • Gate the band gains during frames containing only noise