---
title: "What is audio sample rate? Why speech-to-text uses 16 kHz"
description: "Sample rate is how many audio samples are recorded per second. Why speech models use 16 kHz mono, the file-size math, and how many minutes fit in a 25 MiB upload."
url: https://broadrouter.com/glossary/sample-rate/
lang: en
updated: 2026-11-16
publisher: BroadRouter (a BroadNet site); API by BroadNet Inference, https://inference.broadnet.ai
---

# Sample rate

Updated 2026-11-16

Sample rate is the number of audio samples captured per second, measured in hertz (Hz). Speech recognition models commonly expect 16,000 Hz (16 kHz) mono audio. Together with bit depth and channel count, the sample rate sets how large an uncompressed audio file is per minute.

Also called: sampling rate, sampling frequency, audio sample rate, 16 kHz audio

Sample rate is the first number to check when an audio file is too large or a transcription request is rejected.

## How it works

Digital audio stores the sound wave as a list of numbers, one per sample. At 16,000 Hz there are 16,000 numbers per second per channel. Each number takes 2 bytes at 16-bit depth. Uncompressed WAV size is therefore:

```text
bytes per second = sample rate x bytes per sample x channels
16 kHz, 16-bit, mono:    16,000 x 2 x 1 = 32,000 B/s  -> x 60 = 1,920,000 B per minute
44.1 kHz, 16-bit, stereo: 44,100 x 2 x 2 = 176,400 B/s -> x 60 = 10,584,000 B per minute
```

Speech models are usually trained on 16 kHz audio. OpenAI’s Whisper code sets `SAMPLE_RATE = 16000` and loads every file through ffmpeg with `-ac 1 -ar 16000`. In FFmpeg’s own documentation, `-ar` sets the audio sampling frequency and `-ac` the number of channels.

## Why it matters

-   **File size limits.** A CD-quality stereo WAV is 10,584,000 / 1,920,000 = 5.5 times bigger than 16 kHz mono with no gain for a 16 kHz model.
-   **Upload time.** Smaller files reach the API faster, which matters for voice bots.
-   **Compatibility.** Tools like [voice activity detection](https://broadrouter.com/glossary/vad/) accept only certain rates (WebRTC VAD takes 8, 16, 32 or 48 kHz).
-   **Do not upsample.** Converting an 8 kHz phone call to 16 kHz makes the file twice as big but adds no detail.

## On BroadNet Inference

BroadNet’s transcription endpoint accepts WAV or MP3 up to 26,214,400 bytes (25 MiB) per file. Its docs convert WhatsApp `.ogg` notes and `.m4a` recordings with:

```bash
ffmpeg -i voice-note.ogg -ar 16000 -ac 1 voice-note.wav
```

How much audio fits in one request as 16-bit WAV:

| Format | Bytes per minute | Fits in 26,214,400 bytes |
| --- | --- | --- |
| 16 kHz mono | 1,920,000 | 26,214,400 / 1,920,000 = 13.65 min, about 13 min 39 s |
| 44.1 kHz stereo | 10,584,000 | 2.48 min, about 2 min 29 s |

For anything longer, split into 10-minute segments (600 s x 32,000 B/s = 19,200,000 bytes each), as the [WhatsApp voice notes guide](https://broadrouter.com/guides/transcribe-whatsapp-voice-notes/) shows, or send MP3, which is far smaller per minute. Always set `language`. Billing is per audio minute, not per byte, so the sample rate changes upload size but not the price; check costs in the [speech-to-text cost calculator](https://broadrouter.com/tools/speech-to-text-cost-calculator/). For the bigger picture see [speech-to-text](https://broadrouter.com/glossary/speech-to-text/).

## Questions

### What sample rate should I use for speech-to-text?

16 kHz mono is the common choice. Whisper resamples all input to 16,000 Hz, and BroadNet's docs convert voice notes with ffmpeg -ar 16000 -ac 1.

### How big is one minute of 16 kHz WAV?

For 16-bit mono, 16,000 samples x 2 bytes x 60 seconds = 1,920,000 bytes, about 1.92 MB, plus a small header.

### Does a higher sample rate improve transcription?

Not if the model works at 16 kHz, because it resamples the audio down anyway. A higher rate only makes the upload bigger.

## See also

-   [Speech-to-text](https://broadrouter.com/glossary/speech-to-text/) (term)
-   [Voice activity detection (VAD)](https://broadrouter.com/glossary/vad/) (term)
-   [Speaker diarization](https://broadrouter.com/glossary/diarization/) (term)
-   [Timestamps (speech-to-text)](https://broadrouter.com/glossary/timestamps-asr/) (term)
-   [WAV](https://broadrouter.com/glossary/wav/) (term)
-   [Ogg Opus](https://broadrouter.com/glossary/ogg-opus/) (term)
-   [Transcribe WhatsApp voice notes to text with an API](https://broadrouter.com/guides/transcribe-whatsapp-voice-notes/) (guide)
-   [Arabic speech-to-text API, step by step](https://broadrouter.com/guides/arabic-speech-to-text-api/) (guide)
-   [Voice notes to text for business: WhatsApp and Telegram to tickets](https://broadrouter.com/blog/voice-notes-to-text-for-business/) (article)

Try it on your real workload. New accounts start with $5 of credit for Gemma and transcription. DeepSeek, GLM and Qwen3.8 need a top-up.

[Get started](https://inference.broadnet.ai/account/?utm_source=broadrouter&utm_medium=referral&utm_campaign=glossary-sample-rate&utm_content=page-cta) ·

**Sources**

-   [FFmpeg documentation, -ar and -ac audio options (checked 2026-11-16)](https://ffmpeg.org/ffmpeg.html)
-   [OpenAI Whisper source, audio.py (SAMPLE\_RATE = 16000) (checked 2026-11-16)](https://github.com/openai/whisper/blob/main/whisper/audio.py)
-   [BroadNet Inference API reference, transcription and ffmpeg conversion (checked 2026-11-16)](https://inference.broadnet.ai/docs?utm_source=broadrouter&utm_medium=referral&utm_campaign=glossary-sample-rate&utm_content=sources)
