AlienKevin commited on
Commit
c07efe8
1 Parent(s): b71b2e5

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +3 -0
README.md CHANGED
@@ -42,6 +42,7 @@ Here's a breakdown of the number of recordings under each speaker:
42
  ## Dataset Construction
43
 
44
  1. Gathering
 
45
  We first identified 3,904 common Cantonese syllables based on words.hk's syllable recordings.
46
  The, we ask the speech synthesis APIs to pronounce each of the syllables.
47
  The queries use SSML's phoneme attribute to precisely specify the syllable we want. Here's a sample SSML query that fetches the syllable jyut6:
@@ -52,11 +53,13 @@ The queries use SSML's phoneme attribute to precisely specify the syllable we wa
52
  Apple voices are gathered using jyutping text directly and a native Cantonese ASR system is used to filter out unsupported syllables.
53
 
54
  2. Preprocessing
 
55
  * All audios are converted to 16kHz WAV files
56
  * Peak normalize all audios to -20 dBFS
57
  * Clip silence at the beginning and end (sound below -50 dBFS are deemed silence)
58
 
59
  3. Verification
 
60
  Occassionally, some syllables are not synthesized correctly.
61
  * Apple voices usually renders tone 5 syllables as tone 2: we remove all tone 5 syllables from apple voices
62
  * Microsoft voices prepends consonants like ng, g, and b in front of isolate vowel syllables like aa: we remove all vowel syllables from microsoft voices
 
42
  ## Dataset Construction
43
 
44
  1. Gathering
45
+
46
  We first identified 3,904 common Cantonese syllables based on words.hk's syllable recordings.
47
  The, we ask the speech synthesis APIs to pronounce each of the syllables.
48
  The queries use SSML's phoneme attribute to precisely specify the syllable we want. Here's a sample SSML query that fetches the syllable jyut6:
 
53
  Apple voices are gathered using jyutping text directly and a native Cantonese ASR system is used to filter out unsupported syllables.
54
 
55
  2. Preprocessing
56
+
57
  * All audios are converted to 16kHz WAV files
58
  * Peak normalize all audios to -20 dBFS
59
  * Clip silence at the beginning and end (sound below -50 dBFS are deemed silence)
60
 
61
  3. Verification
62
+
63
  Occassionally, some syllables are not synthesized correctly.
64
  * Apple voices usually renders tone 5 syllables as tone 2: we remove all tone 5 syllables from apple voices
65
  * Microsoft voices prepends consonants like ng, g, and b in front of isolate vowel syllables like aa: we remove all vowel syllables from microsoft voices