About this tool
Sample rate is the number of audio samples recorded each second. Common values include 8, 16, or 22.05 kHz for speech and 44.1 or 48 kHz for music and video. Resampling maps an existing waveform to a new grid for an API, device, or editing-project requirement.
Raising the rate does not add high-frequency detail absent from the source; it only creates more samples. Lowering it requires filtering above the new Nyquist frequency to avoid aliasing. Choose from the destination specification.
How to use it
Check source and destination
Read the current rate and confirm what a speech model, video project, or device requires.
Choose a target rate
Do not raise it without purpose, and consider needed high frequencies when lowering.
Convert and validate
Listen for clarity and inspect file parameters in the receiving application.
Supported range and limits
- Common targets
- 8 kHz (telephony), 16 kHz (speech recognition), 44.1 kHz (CD), 48 kHz (video standard)
- Downsampling
- Discards every frequency above half the new rate; the Nyquist limit; and cannot be undone
- Upsampling
- Adds no information; it interpolates extra samples, and is normally done only to satisfy a downstream format
- Output
- MP3, which supports a limited set of rates; anything outside it lands on the nearest supported rate
- Typical use
- Matching material to what a speech-recognition model or a video project requires
- Where it runs
- FFmpeg compiled to WebAssembly in your browser; the file is never uploaded
When you would use it
Adapting to speech recognition
Standardize a recording to an input requirement such as 16 kHz.
Matching a video project
Convert audio to common 48 kHz to reduce automatic timeline resampling.
Preparing input for speech recognition
Most speech models are trained on 16 kHz mono, so resampling to that spec cuts wasted computation and gives steadier results.
What to know before you start
- Sample rate and bitrate differ: one counts samples, while the other measures encoded data per second.
- Upsampling does not improve real quality, and downsampling limits the highest representable frequency.
- Repeated resampling can accumulate filtering errors; convert once from the source.
Related concepts
- Nyquist frequency
- The theoretical highest representable frequency, approximately half the sample rate.
- resampling
- Converting discrete audio samples from one sampling frequency to another.