What is Sample Rate?
How many times per second audio is captured, in hertz (Hz). Speech data is usually delivered at 16kHz or 48kHz.
16 kHz is the common standard for speech recognition because it covers the frequency range that matters for the human voice. 48 kHz is mostly for music and professional recording.
Downsampling is possible; upsampling cannot invent information that was never captured. So when in doubt, record at the higher rate.
The sample rates you will actually see
Three rates cover most speech data. 8 kHz is the telephone band: call recordings arrive at this rate, and they hold nothing above 4 kHz. 16 kHz is the working standard for speech recognition and reaches 8 kHz, which covers most of what matters for intelligibility. 44.1 and 48 kHz are music and professional recording rates, and they show up mainly when the source is video.
The ceiling is the point to understand: a recording can only hold frequencies up to half its sample rate, so the number decides what is in the file, not just how large it is. Telephone speech loses the top of fricatives such as s and sh, which is why an 8 kHz call recording and a 16 kHz studio recording sound different, and models trained on one do not transfer cleanly across.
Rate is not the only number on the file. Bit depth sets the dynamic range the recording can carry, and the codec decides how much was thrown away to save space: an uncompressed file keeps everything, a compressed one discards detail that cannot be restored. A file can be 16 kHz and still be a poor source if it was heavily compressed.
What to specify when buying
State the rate you need, and match it to the target. If the model runs on telephone audio, buy 8 kHz or downsample; if it runs on wideband speech, buy 16 kHz or higher. Downsampling loses information but is a normal step; upsampling cannot add any. A supplier offering to upsample 8 kHz calls to 16 kHz is changing the number, not the content.
Ask for the rate to be consistent across the batch. Mixed delivery is a common defect: part of a set at 16 kHz, part at 8 kHz, or files whose headers claim one rate while the audio is another. It is cheap to check and expensive to discover during training, so a sample check of the file headers belongs in acceptance testing.
Specify the format as well: mono or stereo, the file container, and whether the original recordings are available. Mono at 16 kHz in an uncompressed format is the safe default for speech. If the originals are higher quality, keeping them means the data can be reused later for a purpose that needs more, such as voice quality work.
Related terms
-
Transcription
Writing down what is said in a recording. It is the core annotation task in any speech dataset.
-
SNR
Signal-to-Noise Ratio, how much louder the speech is than the background noise, measured in decibels (dB).
Keep reading
-
Full glossary
Every term we explain, in one list.
-
How to buy training data
Where these terms actually show up, and which ones change a quote.
-
Help center
How a project runs, from specification to delivery.
Submit a sourcing request
Tell us the language, the hours, and what the data needs to look like. You will get a real number and a real timeline, not a range. If we cannot source it well, we will tell you that instead.
- Pilot batch before the full run, so problems surface early.
- Consent documentation delivered with the data.
- No medical or clinical data. No recorded telephone calls.
- You send the spec Language, hours, recording conditions, and the deadline.
- We reply within two business days A real number and a real timeline, not a range.
- You approve a pilot batch A small run first, so problems surface before the full order.
- Full run, then delivery with documents Audio, transcripts, and the consent record for every speaker.