WAV is the plain format of audio: big, simple, and accepted almost everywhere.
How it works
The Library of Congress describes WAVE as a “wrapper file format” built on RIFF (Resource Interchange File Format), developed by Microsoft and IBM, with a joint specification dated August 1991. A WAV file holds a format description and the audio data. The most common encoding is linear PCM: raw samples, no compression. Standard WAVE is limited to about 4 GB; the RF64 extension exists to go beyond that.
Because PCM is uncompressed, size is simple arithmetic:
bytes per second = sample rate × bytes per sample × channels
For speech-to-text audio at 16 kHz, mono, 16-bit (2 bytes):
| Step | Value |
|---|---|
| 16,000 samples × 2 bytes × 1 channel | 32,000 bytes per second |
| 26,214,400 bytes ÷ 32,000 | 819.2 seconds |
| 819.2 ÷ 60 | about 13.65 minutes (13 min 39 s) |
A stereo 44.1 kHz file of the same length is 44,100 × 2 × 2 = 176,400 bytes a second, more than five times larger, with no gain for transcription.
Why it matters
- Byte caps, not minute caps. Transcription APIs often limit file size. With WAV you can predict exactly how much audio fits.
- No codec surprises. PCM WAV decodes the same everywhere, so it is a safe target when you convert from Ogg Opus, M4A or video.
- Resample down. Higher sample rates and stereo add bytes, not words.
On BroadNet Inference
BroadNet’s transcription endpoint accepts WAV (audio/wav) or MP3 (audio/mpeg), up to 26,214,400 bytes per file. Its docs convert other formats with:
ffmpeg -i voice-note.ogg -ar 16000 -ac 1 voice-note.wav
The ffmpeg docs say -ar sets the sampling frequency, -ac the channel count, and the output format is guessed from the .wav extension. Check the result with ffprobe if you need to confirm 16-bit PCM. Anything longer than about 13 minutes at that setting needs splitting: the audio splitter tool builds the commands, and transcribing long audio files covers joining the text. Billing is per audio minute ($0.0042 on BroadNet Transcribe Arabic), so splitting does not change the audio total.
Common mistakes
- Leaving the original 48 kHz stereo. It fills the cap in under three minutes: 48,000 × 2 × 2 = 192,000 bytes a second.
- Assuming any .wav is PCM. Some recorders write ADPCM or other codecs in a WAV wrapper.