Sampling frequency

Question

A friend of mine who lives in Thailand helps me to translate phrases in Thai. Today he sent me some audiofiles that I opened in Audacity for analyzing. His voice seems to be in some way distorted, or at least not clear. He has been using X-touch tab to record.

Here in Sweden my girlfriend helps me to record a woman's voice and we are using a Samsung Galaxy S3. Here the voice is almost perfect.

What I could see in Audacity is that the files where the sound is not clear have a sample rate of 8000 Hz, compared to the clear soundfiles from samsung which is 44100 Hz.

Does the lower sampling rate produce low quality sound files? I read that if one records speech, a lower sampling rate is ok.

In both cases - the amplitude resolution is the same - that is a 32 bit float.

Answer 1

Lowering the sampling rate affects directly the highest frequency that is captured or played back, in line with Nyquist's sampling theorem, which defines the phenomenon.

The sampling theorem states that:
sampling frequency / 2 = highest audible frequency in the captured/played digital signal

A voice needs around 4-5kHz of bandwidth (around 1kHz-6kHz in audio frequencies) to be tangible. Therefore an 8kHz sample rate (divided by 2 gives 4kHz of bandwidth) suffices and minimizes the need of bandwidth, therefore it needs less storage space and is faster and more efficient (by electrical means) to transfer and thus a low sampling rate such as 8kHz is suitable for mobile phones and other transmission by radio waves.

Answer 2

If the sampling rate was reduced to reduce filesize for transfer, I would suggest looking at bit-rate instead. A lot of podcasts I listen to can get away with 64 kbps mp3s, but I would suggest higher for translation purposes. Still, a 256 kbps mp3 would be a big space saver compared to uncompressed WAV.

@Internet Human, true those are the most important frequencies, but I think 8 kHz is cutting it a bit close. There's still a lot of information being lost. If you're listening to a foreign language or a thick accent, your brain will probably be less likely to fill in the gaps. I would opt for a 22 kHz sampling rate, or record uncompressed and use MP3 encoding.

Answer 3

There is also the question of the quality of hardware (built in MIC and preamp). On my laptop, the built in electret MIC produces horrible, distorted, garbled sound - even when recording to 16bit/44.1kHz. There is all sorts of interference too, like whining from the HD,etc. Bad preamp, dirty power supply. This might be the case with tour friends recording too.

Answer 4

To help with intelligibility, sampling at such low rates would require some emphasis (pre- or de-), though i doubt any of these devices apply it. The other thing to watch out for is aliasing. Here's an interesting paper for more details: http://www.wescottdesign.com/articles/Sampling/sampling.pdf

Answer 5

8 khz sampling rate are not enough for "clear" voice. With 8k you will lose the high frequencies which give clarity to the sound.

44.1khz are needed to capture high-frequencies up to 22khz (human ear typically hear to 16khz) so that will be enough regarding sampling rate.

If you getting the voice files in mp3 format, i suggest 192kbps. Most people will say 128kbps or even 64kbps, but i can hear the difference between 192 and 128 kbps, even though my hearing is suffering a bit. 192kbps should be enough to have clear voice.