3 ms·
If you make your training set the output of Google's speech recognition algo from YT there will be lots of label noise. It's better to go with movies and audio
by visarga 10y ago
If you make your training set the output of Google's speech recognition algo from YT there will be lots of label noise.
It's better to go with movies and audiobooks because the text 'transcription' is human generated. Aligning known text to audio is much more accurate than speech recognition.