Здравствуйте Николай,
подскажите пожалуйста, есть ли какойто установленный порог громкости для
распознаваемой речи? спрашиваю для того, чтобы понять, как можно распознавать
более тихую речь или речь человека, находящегося на бОльшем расстоянии от
микрофона, чем если бы у него были надеты наушники.. если единственный способ
- это усиливать громкость записанного куска речи специальными программами - то
есть ли в pocketsphinx такая встроенная возможность?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi Nikolay,
advice please, if there is a fixed volume level for recognized speech? i ask
to solve a problem when the speech of a person is quiet or the person is quite
far from the mic.. if the only method is just to gain the recorded chop of
speech in special programs (such as audacity or smth) - tell please, if
pocketsphinx provides some feature like this
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
ASR systems are not really affected by energy level. They have it normalized.
Unless you make a volume too low to loose information due to quantization
errors the accuracy should be the same.
The biggest problem with long-distance speech recognition is a reverberation
and a noise cancellation, not the volume level. You can find more information
about reverberation compensation in papers.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Здравствуйте Николай,
подскажите пожалуйста, есть ли какойто установленный порог громкости для
распознаваемой речи? спрашиваю для того, чтобы понять, как можно распознавать
более тихую речь или речь человека, находящегося на бОльшем расстоянии от
микрофона, чем если бы у него были надеты наушники.. если единственный способ
- это усиливать громкость записанного куска речи специальными программами - то
есть ли в pocketsphinx такая встроенная возможность?
Hi Nikolay,
advice please, if there is a fixed volume level for recognized speech? i ask
to solve a problem when the speech of a person is quiet or the person is quite
far from the mic.. if the only method is just to gain the recorded chop of
speech in special programs (such as audacity or smth) - tell please, if
pocketsphinx provides some feature like this
ASR systems are not really affected by energy level. They have it normalized.
Unless you make a volume too low to loose information due to quantization
errors the accuracy should be the same.
The biggest problem with long-distance speech recognition is a reverberation
and a noise cancellation, not the volume level. You can find more information
about reverberation compensation in papers.
Thanks very much, i will proceed.