How does the same voice look under different conditions?
Each row below is the same spoken sentence passed through every noise / channel condition. The images are mel spectrograms — bright colors mean strong sound at that frequency (vertical) at that moment (horizontal).
Watch how high frequencies disappear in the phone conditions, how noise fills the background, and how quiet segments get swallowed in the low-volume version.
Example 1 — file_0025_conv_004_2a
"እሽ ብርሃን ለድርጅቶች የቀረቡ አራት አይነት የጽሁፍ መልእክት ጥቅሎች አሉን በቀን ስንት መልእክት ለመላክ አስበዋል"
Clean Studio
Ambient Room Noise
Harsh Environment
Distant / Low-Volume Mic
Wideband Phone (VoLTE)
Randomized Phone Channel
Narrowband 2G + Packet Loss
Example 2 — file_0381_conv_002_3u
"የድምጽ እና የጽሁፍ መልእክት ብቻ ቢሆን ይሻለኛል እንደት ነው የሚከፈተው"
Clean Studio
Ambient Room Noise
Harsh Environment
Distant / Low-Volume Mic
Wideband Phone (VoLTE)
Randomized Phone Channel
Narrowband 2G + Packet Loss
Example 3 — file_1357_conv_001_3u
"እሽ አሁን እንደት ነው የምጠቀመው አድስ ቁጥር አልደረሰኝም"
Clean Studio
Ambient Room Noise
Harsh Environment
Distant / Low-Volume Mic
Wideband Phone (VoLTE)
Randomized Phone Channel
Narrowband 2G + Packet Loss
Example 4 — file_0737_conv_004_6a
"አይደለም አገልግሎቱ ክፍት ሆኖ ቢቆይም ካልተጠቀሙበት ሂሳብ አይቆርጥም ነገር ግን ለጥንቃቄ ማቋረጡ ጥሩ ነው"
Clean Studio
Ambient Room Noise
Harsh Environment
Distant / Low-Volume Mic
Wideband Phone (VoLTE)
Randomized Phone Channel
Narrowband 2G + Packet Loss
What you're seeing. Speech energy concentrates in a few frequency bands (formants).
Phone / narrowband codecs chop off everything above ~3–4 kHz, so the top of the spectrogram goes dark.
Heavy background noise adds a haze everywhere. Low-audio simply dims the whole picture.