Department of Computer Science, The University of Hong Kong

COMP7503B
Multimedia Technologies

Deadline: 08 Mar 2024, 11:55pm

Preamble:

In this assignment, you are required to analyze phoneme patterns based on spectral
analysis technique. In order to perform spectral analysis, you can do so based on the following
mobile apps on Android/iOS platform.

Android Platform:
Recommended: Spectral Audio Analyzer
https://play.google.com/store/apps/details?id=radonsoft.net.spectralview&hl=en

iOS Platform:
Recommended: SpectrumView
https://itunes.apple.com/us/app/spectrumview/id472662922?mt=8

After you install these apps, you should grant the permission for the App to access the
microphone on your mobile device. You should assume default settings of the app, except for
sampling rate has to be set to allow the capture of 22 kHz audio signal.

Questions:

1. Launch the app on your mobile device in a quiet environment, speak the sentence “Peter
buttered the burnt toast”. Take the screenshot of the spectrogram, and indicate the presence
of the stop consonants and voiced phonemes in the spectrogram.

2. Repeat question 1 but this time with the sentence changed to “I owe you a Yo-Yo”.

3. To implement speech to text algorithm, it is important to do word segmentation during

spectrogram analysis. From what you observed in question 1 & 2, explain why it is not
easy to do word segmentation based on silence detection.

4. Explain, by way of illustrating a number of sample spectrograms, why it is necessary to use

phonetic alphabet (i..e A- Alpha, B - Beta, C - Charlie, …) to ensure correct exchange of
letters over voice messages by radio or telephone.

=== End ===