Skip to content
orukSign up
Speech-to-textVoice agentsModelsPricingResearchAPI docsQuickstartSDKsMCP serverIntegrationsBenchmarksStatusLog in

What 8 kHz μ-law does to telephone audio

Compare an original 16 kHz recording with an 8 kHz G.711 μ-law round trip. Listen for consonant detail and inspect the frequency band lost in the telephone version.

Listen and inspect

Play the original “Don’t forget a jacket,” then choose the telephone-codec version. Watch the region above 4 kHz. The second file was downsampled and μ-law encoded, then decoded to linear PCM for playback.

16 kHz original · CREMA-D

Local files play in your browser and stay on your device.

Original recording from CREMA-D, file 1080_DFA_HAP_XX. Its source and hash, database license and contents license apply to the audio used here.

What to listen for

The 8 kHz sampling rate limits representable frequencies to below 4 kHz. μ-law also quantizes amplitude nonlinearly. Converting back to 16 kHz for a speech API changes the representation without recovering the discarded band.

Twilio · Media Streams WebSocket audio format

How to test it

  1. Decode base64 media payloads to their original μ-law bytes. Decode to linear PCM before resampling and keep each caller track separate.
  2. Replay a clean source, a downsample-only control and its μ-law round trip through the same model. Compare word errors and the score shifts on each matched utterance.
  3. Record channel direction, audio format, packet gaps and resampler settings. This file pair models codec conversion only. Real telephone paths can add filtering, packet loss, echo cancellation and noise.

Run the Twilio-to-Oruk audio bridge

Try it on your own audio.

The demo needs no account. The trial needs a card.