Speech & Audio Quiz
Five questions spanning the course — sampling and the Nyquist limit, spectrograms and mel features, the alignment problem, CTC, and Whisper-style ASR.
Exercise
You need to capture audio content up to 10 kHz. What minimum sampling rate does the Nyquist theorem require?
Need a hint?
Sampling rate must be at least twice the highest frequency.
Exercise
Choosing the STFT window length involves a fundamental trade-off. What is it?
Need a hint?
Resolution in time vs resolution in frequency.
Exercise
What are MFCCs (Mel-Frequency Cepstral Coefficients)?
Need a hint?
A compact transform of the log-mel energies.
Exercise
Compared to an attention-based sequence-to-sequence ASR model, what is a key advantage of CTC (and transducer/RNN-T) models?
Need a hint?
Which can produce output as you speak?
Exercise
What architecture does Whisper use, and what does it take as input?
Need a hint?
A familiar transformer setup, fed a feature from Lesson 1.
Exercise
Converting a raw waveform into a log-mel spectrogram before training an ASR model is a decision in which slot of the ML project loop?
Need a hint?
It's about the representation the model consumes, upstream of the architecture.
Exercise
Transfer test — a technique this course did NOT teach: 8-bit quantization of a trained model's weights to make inference cheaper. Which slot of the ML project loop does it serve?
Need a hint?
It's applied after training to make the deployed model cheaper to run.