Description

Assignment 11
Total: 60 Points

General Instruction
• Submit your work in the Dropbox folder via BeachBoard. (Not email or in class)
• Submit the separate files as they are. (no zip file)

1. Evaluate the performance of Google Web Speech API.
(a) Find the speech.py.
(b) Read How Speech Recognition Works.txt which includes 25 sentences, and record your speech as separate WAV(PCM) files using the nomenclature ‘Group ID-Sent#.wav’. For instance, 01-Sent01.wav. I recommend you to use Audacity to record and edit your speech.
(d) (5 points) Complete the read original method that imports the ‘How Speech Recognition Works.txt’ into a list of strings, self.original, in the order of the sentence number.
(e) (20 points) Refer The Ultimate Guide To Speech Recognition With Python – Real Python.mhtml, and complete the conv audio method that converts audio files into a list of strings, self.recognized. The method should convert all audio files (.wav) in the folder, inDir, in the order of Sent#.
(f) (10 points) Complete the comp string method that compares two lists of strings, self.original and self.recognized, and calculates the similarities of two strings by using Levenshtein Distance (LD). The normalized Levenshtein distance (NLD) is defined by
LD
NLD =
max(# of words in the original,# of words in the recognized)
(You need to convert the strings into the lists of words. For instance, ‘I love AI’ to [‘I’, ‘love’, ‘AI’]. Please refer this site.) This method should store the separate distances (NLD) in self.distances in the order of Sent#.
(g) Visit ‘BeachBoard – Discussions – Lab – How Speech Recognition Works’ again, anddownload all submissions of other groups. Since we have 23 groups, (ideally) you will have 23 audio files for each sentence.
CECS 451 Assignment 11 – Page 2 of 2

(i) Submit your speech.ipynb.

Figure 1: An example of the box-and-whisker plot of normalize distances by sentences. Please note that the plot is generated using random numbers, which will be different than your result.