Resample to 8 kHz
Drop an audio file and export it at 8 kHz as a WAV. This is the narrowband rate telephony runs at, and it is the lowest rate in common use that still carries understandable speech.
The 4 kHz band
Half of 8 kHz is 4 kHz, and that is the ceiling. Telephone systems carry roughly 300 Hz to 3.4 kHz, a range settled on in the analogue era as the least that keeps a voice recognisable and a conversation workable. The digital rate was chosen to hold it.
Everything above the ceiling is filtered out before the rate drops, so it cannot fold back into the voice band as an aliased tone.
What intelligibility costs
Speech stays understandable but loses precision. The cues separating f from s, or th from f, sit largely above 4 kHz, so those consonants blur into each other. This is why reading out a code or a name over a phone line often needs a spelling alphabet.
Music does not survive the conversion in any useful form.
When to pick it
- Preparing audio for a telephony or IVR system that requires narrowband input.
- Matching an existing 8 kHz recording so a set of files shares one rate.
- Deliberately producing a telephone effect, where the band limit is the point.
- Cutting file size to the minimum where only intelligibility matters.
Not the rate for transcription
Speech models are trained at 16 kHz and rely on detail between 4 and 8 kHz that this rate removes. Converting to 8 kHz before transcription lowers accuracy. Use 16 kHz unless the receiving system demands narrowband.
For a different rate, a mono downmix, or another output format, use the full Sample Rate Converter.